diff --git a/studio/backend/core/training/trainer.py b/studio/backend/core/training/trainer.py
index 5f504bbdf4..2324916236 100644
--- a/studio/backend/core/training/trainer.py
+++ b/studio/backend/core/training/trainer.py
@@ -81,8 +81,8 @@ class TrainingProgress:
     epoch: float = 0
     step: int = 0
     total_steps: int = 0
-    loss: float = 0.0
-    learning_rate: float = 0.0
+    loss: Optional[float] = None
+    learning_rate: Optional[float] = None
     is_training: bool = False
     is_completed: bool = False
     error: Optional[str] = None
@@ -244,7 +244,7 @@ class UnslothTrainer:
             def on_log(self, args, state, control, logs = None, **kwargs):
                 if not logs:
                     return
-                loss_value = logs.get("loss", logs.get("train_loss", 0.0))
+                loss_value = logs.get("loss", logs.get("train_loss", None))
                 current_step = state.global_step
                 grad_norm = logs.get("grad_norm", None)
 
@@ -268,7 +268,7 @@ class UnslothTrainer:
                     step = current_step,
                     epoch = round(state.epoch, 2) if state.epoch else 0,
                     loss = loss_value,
-                    learning_rate = logs.get("learning_rate", 0.0),
+                    learning_rate = logs.get("learning_rate", None),
                     elapsed_seconds = elapsed_seconds,
                     eta_seconds = eta_seconds,
                     grad_norm = grad_norm,
diff --git a/studio/backend/core/training/training.py b/studio/backend/core/training/training.py
index 9626f9df2e..4439e4e173 100644
--- a/studio/backend/core/training/training.py
+++ b/studio/backend/core/training/training.py
@@ -14,12 +14,14 @@ worker's mp.Queue, and exposes the same API surface to routes/training.py.
 Pattern follows core/data_recipe/jobs/manager.py.
 """
 
+import json as _json
 import math
 import multiprocessing as mp
 import queue
 import threading
 import time
 import structlog
+from datetime import datetime, timezone
 from loggers import get_logger
 from dataclasses import dataclass, field
 from pathlib import Path
@@ -44,8 +46,8 @@ class TrainingProgress:
     epoch: float = 0
     step: int = 0
     total_steps: int = 0
-    loss: float = 0.0
-    learning_rate: float = 0.0
+    loss: Optional[float] = None
+    learning_rate: Optional[float] = None
     is_training: bool = False
     is_completed: bool = False
     error: Optional[str] = None
@@ -63,6 +65,8 @@ class TrainingBackend:
     Launches a fresh subprocess per training job, communicates via mp.Queue.
     """
 
+    FLUSH_THRESHOLD: int = 10
+
     def __init__(self):
         # Subprocess state
         self._proc: Optional[mp.Process] = None
@@ -91,13 +95,21 @@ class TrainingBackend:
         self.current_job_id: Optional[str] = None
         self._output_dir: Optional[str] = None
 
+        # DB persistence
+        self._metric_buffer: list[dict] = []
+        self._run_finalized: bool = False
+        self._db_run_created: bool = False
+        self._db_total_steps_set: bool = False
+        self._db_config: Optional[dict] = None
+        self._db_started_at: Optional[str] = None
+
         logger.info("TrainingBackend initialized (subprocess mode)")
 
     # ------------------------------------------------------------------
     # Public API (called by routes/training.py)
     # ------------------------------------------------------------------
 
-    def start_training(self, **kwargs) -> bool:
+    def start_training(self, job_id: str, **kwargs) -> bool:
         """Spawn a subprocess to run the full training pipeline.
 
         All kwargs are serialized into a config dict and sent to the worker.
@@ -108,30 +120,16 @@ class TrainingBackend:
                 logger.warning("Training subprocess already running")
                 return False
 
-        # Join prior pump thread to prevent it from consuming events
-        # from the new job's queue (it reads self._event_queue dynamically).
+        # Join prior pump thread — refuse to start if it won't die
         if self._pump_thread is not None and self._pump_thread.is_alive():
             self._pump_thread.join(timeout = 5.0)
             if self._pump_thread.is_alive():
-                logger.warning("Previous pump thread did not exit within 5s")
+                logger.warning(
+                    "Previous pump thread did not exit within 5s — refusing to start"
+                )
+                return False
         self._pump_thread = None
 
-        # Reset state
-        self._should_stop = False
-        self._cancel_requested = False
-        self._progress = TrainingProgress(
-            is_training = True, status_message = "Initializing training..."
-        )
-        self.loss_history.clear()
-        self.lr_history.clear()
-        self.step_history.clear()
-        self.grad_norm_history.clear()
-        self.grad_norm_step_history.clear()
-        self.eval_loss_history.clear()
-        self.eval_step_history.clear()
-        self.eval_enabled = False
-        self._output_dir = None
-
         # Build config dict for the subprocess
         config = {
             "model_name": kwargs["model_name"],
@@ -193,23 +191,62 @@ class TrainingBackend:
         if config["training_type"] != "LoRA/QLoRA":
             config["load_in_4bit"] = False
 
-        # Spawn subprocess
+        # Spawn subprocess — use locals so state is untouched on failure
         from .worker import run_training_process
 
-        self._event_queue = _CTX.Queue()
-        self._stop_queue = _CTX.Queue()
+        event_queue = _CTX.Queue()
+        stop_queue = _CTX.Queue()
 
-        self._proc = _CTX.Process(
+        proc = _CTX.Process(
             target = run_training_process,
             kwargs = {
-                "event_queue": self._event_queue,
-                "stop_queue": self._stop_queue,
+                "event_queue": event_queue,
+                "stop_queue": stop_queue,
                 "config": config,
             },
             daemon = True,
         )
-        self._proc.start()
-        logger.info("Training subprocess started (pid=%s)", self._proc.pid)
+        try:
+            proc.start()
+        except Exception:
+            logger.error("Failed to start training subprocess", exc_info = True)
+            return False
+
+        logger.info("Training subprocess started (pid=%s)", proc.pid)
+
+        # Reset state — safe because old pump thread is confirmed dead
+        # and proc.start() succeeded
+        self.current_job_id = job_id
+        self._should_stop = False
+        self._cancel_requested = False
+        self._progress = TrainingProgress(
+            is_training = True, status_message = "Initializing training..."
+        )
+        self.loss_history.clear()
+        self.lr_history.clear()
+        self.step_history.clear()
+        self.grad_norm_history.clear()
+        self.grad_norm_step_history.clear()
+        self.eval_loss_history.clear()
+        self.eval_step_history.clear()
+        self.eval_enabled = False
+        self._output_dir = None
+        self._metric_buffer.clear()
+        self._run_finalized = False
+        self._db_run_created = False
+        self._db_total_steps_set = False
+        self._db_config = {
+            k: v for k, v in config.items() if k not in {"hf_token", "wandb_token"}
+        }
+        self._db_started_at = datetime.now(timezone.utc).isoformat()
+
+        # Assign subprocess handles after state reset
+        self._event_queue = event_queue
+        self._stop_queue = stop_queue
+        self._proc = proc
+
+        # Eagerly create DB run row so the run appears in history during model loading
+        self._ensure_db_run_created()
 
         # Start event pump thread
         self._pump_thread = threading.Thread(target = self._pump_loop, daemon = True)
@@ -252,6 +289,11 @@ class TrainingBackend:
                 proc.kill()
                 proc.join(timeout = 2.0)
 
+        # Wait for pump thread to finish DB finalization before returning
+        # (8s covers SQLite's default 5s lock timeout plus execution overhead)
+        if self._pump_thread is not None and self._pump_thread.is_alive():
+            self._pump_thread.join(timeout = 8.0)
+
     def is_training_active(self) -> bool:
         """Check if training is currently active."""
         with self._lock:
@@ -389,20 +431,54 @@ class TrainingBackend:
                             self._progress.error
                             or "Training process exited unexpectedly"
                         )
+
+            self._ensure_db_run_created()
+            self._finalize_run_in_db(
+                status = "stopped" if self._should_stop else "error",
+                error_message = None
+                if self._should_stop
+                else "Training process terminated unexpectedly",
+            )
             return
 
     def _handle_event(self, event: dict) -> None:
-        """Apply a subprocess event to local state."""
+        """Apply a subprocess event to local state.
+
+        State updates happen inside self._lock; DB I/O happens after
+        releasing it so status-polling API endpoints are never blocked
+        by slow SQLite writes.
+        """
         etype = event.get("type")
+        db_action: Optional[str] = None
+        db_action_kwargs: dict = {}
 
         with self._lock:
             if etype == "progress":
                 self._progress.step = event.get("step", self._progress.step)
                 self._progress.epoch = event.get("epoch", self._progress.epoch)
-                self._progress.loss = event.get("loss", self._progress.loss)
-                self._progress.learning_rate = event.get(
-                    "learning_rate", self._progress.learning_rate
-                )
+                # loss/lr are sanitized below; update progress after coercion
+                _raw_loss = event.get("loss")
+                _raw_lr = event.get("learning_rate")
+                try:
+                    _safe_loss = float(_raw_loss) if _raw_loss is not None else None
+                except (TypeError, ValueError):
+                    logger.debug("Could not convert loss to float: %s", _raw_loss)
+                    _safe_loss = None
+                if _safe_loss is not None and not math.isfinite(_safe_loss):
+                    _safe_loss = None
+                try:
+                    _safe_lr = float(_raw_lr) if _raw_lr is not None else None
+                except (TypeError, ValueError):
+                    logger.debug(
+                        "Could not convert learning_rate to float: %s", _raw_lr
+                    )
+                    _safe_lr = None
+                if _safe_lr is not None and not math.isfinite(_safe_lr):
+                    _safe_lr = None
+                if _safe_loss is not None:
+                    self._progress.loss = _safe_loss
+                if _safe_lr is not None:
+                    self._progress.learning_rate = _safe_lr
                 self._progress.total_steps = event.get(
                     "total_steps", self._progress.total_steps
                 )
@@ -416,30 +492,85 @@ class TrainingBackend:
                 if status:
                     self._progress.status_message = status
 
-                # Update metric histories
+                # Update metric histories — reuse sanitized values from above
                 step = event.get("step", 0)
-                loss = event.get("loss", 0.0)
-                lr = event.get("learning_rate", 0.0)
-                if step >= 0 and loss > 0:
+                loss = _safe_loss
+                lr = _safe_lr
+                if step > 0 and loss is not None:
                     self.loss_history.append(loss)
-                    self.lr_history.append(lr)
+                    self.lr_history.append(lr if lr is not None else 0.0)
                     self.step_history.append(step)
 
                 grad_norm = event.get("grad_norm")
+                gn = None
                 if grad_norm is not None:
                     try:
                         gn = float(grad_norm)
                     except (TypeError, ValueError):
                         gn = None
-                    if gn is not None and math.isfinite(gn):
+                    if step > 0 and gn is not None and math.isfinite(gn):
                         self.grad_norm_history.append(gn)
                         self.grad_norm_step_history.append(step)
+                    else:
+                        gn = None
 
                 eval_loss = event.get("eval_loss")
                 if eval_loss is not None:
-                    self.eval_loss_history.append(eval_loss)
-                    self.eval_step_history.append(step)
-                    self.eval_enabled = True
+                    try:
+                        eval_loss = float(eval_loss)
+                    except (TypeError, ValueError):
+                        logger.debug(
+                            "Could not convert eval_loss to float: %s", eval_loss
+                        )
+                        eval_loss = None
+                    if step > 0 and eval_loss is not None and math.isfinite(eval_loss):
+                        self.eval_loss_history.append(eval_loss)
+                        self.eval_step_history.append(step)
+                        self.eval_enabled = True
+                    else:
+                        eval_loss = None
+
+                # Buffer metric for DB flush (loss/lr already sanitized above)
+                self._metric_buffer.append(
+                    {
+                        "step": step,
+                        "loss": loss,
+                        "learning_rate": lr,
+                        "grad_norm": gn,
+                        "eval_loss": eval_loss,
+                        "epoch": event.get("epoch"),
+                        "num_tokens": event.get("num_tokens"),
+                        "elapsed_seconds": event.get("elapsed_seconds"),
+                    }
+                )
+
+                # Decide which DB action to take after releasing the lock
+                if not self._db_run_created and self.current_job_id and self._db_config:
+                    db_action = "create_run"
+                    db_action_kwargs = {
+                        "job_id": self.current_job_id,
+                        "model_name": self._db_config["model_name"],
+                        "dataset_name": self._db_config.get("hf_dataset")
+                        or next(
+                            iter(self._db_config.get("local_datasets") or []), "unknown"
+                        ),
+                        "config_json": _json.dumps(self._db_config),
+                        "started_at": self._db_started_at
+                        or datetime.now(timezone.utc).isoformat(),
+                        "total_steps": event.get("total_steps"),
+                    }
+                elif (
+                    event.get("total_steps")
+                    and self._db_run_created
+                    and not self._db_total_steps_set
+                ):
+                    db_action = "update_total_steps"
+                    db_action_kwargs = {
+                        "job_id": self.current_job_id,
+                        "total_steps": event["total_steps"],
+                    }
+                elif len(self._metric_buffer) >= self.FLUSH_THRESHOLD:
+                    db_action = "flush"
 
             elif etype == "eval_configured":
                 self.eval_enabled = True
@@ -454,6 +585,14 @@ class TrainingBackend:
                 self._output_dir = event.get("output_dir")
                 msg = event.get("status_message", "Training completed")
                 self._progress.status_message = msg
+                if not self._db_run_created and self.current_job_id and self._db_config:
+                    db_action = "create_and_finalize"
+                else:
+                    db_action = "finalize"
+                db_action_kwargs = {
+                    "status": "stopped" if self._should_stop else "completed",
+                    "output_dir": self._output_dir,
+                }
 
             elif etype == "error":
                 self._progress.is_training = False
@@ -462,6 +601,149 @@ class TrainingBackend:
                 stack = event.get("stack", "")
                 if stack:
                     logger.error("Stack trace:\n%s", stack)
+                if not self._db_run_created and self.current_job_id and self._db_config:
+                    db_action = "create_and_finalize"
+                else:
+                    db_action = "finalize"
+                db_action_kwargs = {
+                    "status": "stopped" if self._should_stop else "error",
+                    "error_message": event.get("error", "Unknown error"),
+                }
+
+        # --- DB I/O outside the lock ---
+        if db_action == "create_run":
+            try:
+                from storage.studio_db import create_run
+
+                create_run(
+                    id = db_action_kwargs["job_id"],
+                    model_name = db_action_kwargs["model_name"],
+                    dataset_name = db_action_kwargs["dataset_name"],
+                    config_json = db_action_kwargs["config_json"],
+                    started_at = db_action_kwargs["started_at"],
+                    total_steps = db_action_kwargs["total_steps"],
+                )
+                self._db_run_created = True
+                if db_action_kwargs["total_steps"]:
+                    self._db_total_steps_set = True
+            except Exception:
+                logger.warning("Failed to create DB run record", exc_info = True)
+        elif db_action == "create_and_finalize":
+            self._ensure_db_run_created()
+            self._finalize_run_in_db(**db_action_kwargs)
+        elif db_action == "update_total_steps":
+            try:
+                from storage.studio_db import update_run_total_steps
+
+                update_run_total_steps(
+                    db_action_kwargs["job_id"], db_action_kwargs["total_steps"]
+                )
+                self._db_total_steps_set = True
+            except Exception:
+                logger.warning("Failed to update total_steps in DB", exc_info = True)
+        elif db_action == "flush":
+            self._flush_metrics_to_db()
+        elif db_action == "finalize":
+            self._finalize_run_in_db(**db_action_kwargs)
+
+    def _ensure_db_run_created(self) -> None:
+        """Create the DB row if it doesn't exist yet. Called outside the lock."""
+        if self._db_run_created or not self.current_job_id or not self._db_config:
+            return
+        try:
+            from storage.studio_db import create_run
+
+            dataset_name = self._db_config.get("hf_dataset") or next(
+                iter(self._db_config.get("local_datasets") or []), "unknown"
+            )
+            create_run(
+                id = self.current_job_id,
+                model_name = self._db_config["model_name"],
+                dataset_name = dataset_name,
+                config_json = _json.dumps(self._db_config),
+                started_at = self._db_started_at
+                or datetime.now(timezone.utc).isoformat(),
+                total_steps = self._progress.total_steps or None,
+            )
+            self._db_run_created = True
+        except Exception:
+            logger.warning(
+                "Failed to create DB run record for early failure", exc_info = True
+            )
+
+    def _finalize_run_in_db(
+        self,
+        status: str,
+        error_message: Optional[str] = None,
+        output_dir: Optional[str] = None,
+    ) -> None:
+        """Flush remaining metrics and mark a run as finished in the DB."""
+        if not self.current_job_id or not self._db_run_created or self._run_finalized:
+            return
+        self._flush_metrics_to_db()
+        try:
+            from storage.studio_db import finish_run
+            from utils.downsample import downsample
+
+            sparkline = downsample(self.loss_history, 50)
+            finish_run(
+                id = self.current_job_id,
+                status = status,
+                ended_at = datetime.now(timezone.utc).isoformat(),
+                final_step = self._progress.step,
+                final_loss = self._progress.loss
+                if (
+                    self._progress.loss is not None
+                    and math.isfinite(self._progress.loss)
+                )
+                else None,
+                duration_seconds = self._progress.elapsed_seconds,
+                loss_sparkline = _json.dumps(sparkline),
+                output_dir = output_dir,
+                error_message = error_message,
+            )
+            self._run_finalized = True
+        except Exception:
+            logger.warning(
+                "Failed to finalize run in DB (status=%s)", status, exc_info = True
+            )
+
+    def _flush_metrics_to_db(self) -> None:
+        """Flush buffered metrics to the database and update live progress."""
+        if (
+            not self._metric_buffer
+            or not self.current_job_id
+            or not self._db_run_created
+        ):
+            return
+        # Cap buffer to prevent unbounded memory growth
+        if len(self._metric_buffer) > 500:
+            logger.warning(
+                "Metric buffer exceeded 500 entries (%d) — trimming oldest",
+                len(self._metric_buffer),
+            )
+            self._metric_buffer = self._metric_buffer[-500:]
+        # Snapshot before insert so metrics arriving during the write are preserved
+        batch = list(self._metric_buffer)
+        try:
+            from storage.studio_db import insert_metrics_batch, update_run_progress
+
+            insert_metrics_batch(self.current_job_id, batch)
+            del self._metric_buffer[: len(batch)]
+            update_run_progress(
+                id = self.current_job_id,
+                step = self._progress.step,
+                loss = self._progress.loss
+                if (
+                    self._progress.loss is not None
+                    and math.isfinite(self._progress.loss)
+                )
+                else None,
+                duration_seconds = self._progress.elapsed_seconds,
+            )
+        except Exception:
+            # Leave buffer intact for retry on next flush
+            logger.warning("Failed to flush metrics to DB", exc_info = True)
 
     @staticmethod
     def _read_queue(q: Any, timeout_sec: float) -> Optional[dict]:
@@ -561,11 +843,13 @@ class TrainingBackend:
             if progress.error:
                 title = f"Error: {progress.error}"
             elif progress.is_completed:
-                title = f"Training completed! Final loss: {progress.loss:.4f}"
+                loss_str = f"{progress.loss:.4f}" if progress.loss is not None else "--"
+                title = f"Training completed! Final loss: {loss_str}"
             elif progress.status_message:
                 title = progress.status_message
             elif progress.step > 0:
-                title = f"Epoch: {progress.epoch} | Step: {progress.step}/{progress.total_steps} | Loss: {progress.loss:.4f}"
+                loss_str = f"{progress.loss:.4f}" if progress.loss is not None else "--"
+                title = f"Epoch: {progress.epoch} | Step: {progress.step}/{progress.total_steps} | Loss: {loss_str}"
             else:
                 title = "Training Loss"
 
diff --git a/studio/backend/core/training/worker.py b/studio/backend/core/training/worker.py
index ccd805b7ac..d06dd6d358 100644
--- a/studio/backend/core/training/worker.py
+++ b/studio/backend/core/training/worker.py
@@ -242,7 +242,7 @@ def run_training_process(
 
     # Wire up progress callback → event_queue
     def _on_progress(progress: TrainingProgress):
-        has_train_loss = progress.step >= 0 and progress.loss > 0
+        has_train_loss = progress.step > 0 and progress.loss is not None
         has_eval_loss = progress.eval_loss is not None
         if has_train_loss or has_eval_loss:
             event_queue.put(
@@ -918,7 +918,7 @@ def _run_embedding_training(event_queue: Any, stop_queue: Any, config: dict) ->
         def on_log(self, args, state, control, logs = None, **kwargs):
             if not logs:
                 return
-            loss_value = logs.get("loss", logs.get("train_loss", 0.0))
+            loss_value = logs.get("loss", logs.get("train_loss", None))
             current_step = state.global_step
 
             elapsed = time.time() - training_start_time
@@ -934,7 +934,7 @@ def _run_embedding_training(event_queue: Any, stop_queue: Any, config: dict) ->
                     "step": current_step,
                     "epoch": round(state.epoch, 2) if state.epoch else 0,
                     "loss": loss_value,
-                    "learning_rate": logs.get("learning_rate", 0.0),
+                    "learning_rate": logs.get("learning_rate", None),
                     "total_steps": total_steps,
                     "elapsed_seconds": elapsed,
                     "eta_seconds": eta,
diff --git a/studio/backend/main.py b/studio/backend/main.py
index 7134c5a783..5e647f6312 100644
--- a/studio/backend/main.py
+++ b/studio/backend/main.py
@@ -49,6 +49,7 @@ from routes import (
     export_router,
     inference_router,
     models_router,
+    training_history_router,
     training_router,
 )
 from auth import storage
@@ -73,6 +74,17 @@ async def lifespan(app: FastAPI):
     # Detect hardware first — sets DEVICE global used everywhere
     detect_hardware()
 
+    from storage.studio_db import cleanup_orphaned_runs
+
+    try:
+        cleanup_orphaned_runs()
+    except Exception as exc:
+        import structlog
+
+        structlog.get_logger(__name__).warning(
+            "cleanup_orphaned_runs failed at startup: %s", exc
+        )
+
     # Pre-cache the helper GGUF model for LLM-assisted dataset detection.
     # Runs in a background thread so it doesn't block server startup.
     import threading
@@ -149,6 +161,9 @@ app.include_router(inference_router, prefix = "/v1", tags = ["openai-compat"])
 app.include_router(datasets_router, prefix = "/api/datasets", tags = ["datasets"])
 app.include_router(data_recipe_router, prefix = "/api/data-recipe", tags = ["data-recipe"])
 app.include_router(export_router, prefix = "/api/export", tags = ["export"])
+app.include_router(
+    training_history_router, prefix = "/api/train", tags = ["training-history"]
+)
 
 
 # ============ Health and System Endpoints ============
diff --git a/studio/backend/models/__init__.py b/studio/backend/models/__init__.py
index 11cf215f54..a4fbbbe6ee 100644
--- a/studio/backend/models/__init__.py
+++ b/studio/backend/models/__init__.py
@@ -10,6 +10,11 @@ from .training import (
     TrainingJobResponse,
     TrainingStatus,
     TrainingProgress,
+    TrainingRunSummary,
+    TrainingRunListResponse,
+    TrainingRunMetrics,
+    TrainingRunDetailResponse,
+    TrainingRunDeleteResponse,
 )
 from .models import (
     CheckpointInfo,
@@ -71,6 +76,11 @@ __all__ = [
     "TrainingJobResponse",
     "TrainingStatus",
     "TrainingProgress",
+    "TrainingRunSummary",
+    "TrainingRunListResponse",
+    "TrainingRunMetrics",
+    "TrainingRunDetailResponse",
+    "TrainingRunDeleteResponse",
     # Model management schemas
     "ModelDetails",
     "LocalModelInfo",
diff --git a/studio/backend/models/training.py b/studio/backend/models/training.py
index 342e44cc09..68791aa7a8 100644
--- a/studio/backend/models/training.py
+++ b/studio/backend/models/training.py
@@ -177,8 +177,8 @@ class TrainingProgress(BaseModel):
     job_id: str = Field(..., description = "Training job identifier")
     step: int = Field(..., description = "Current training step")
     total_steps: int = Field(..., description = "Total training steps")
-    loss: float = Field(..., description = "Current loss value")
-    learning_rate: float = Field(..., description = "Current learning rate")
+    loss: Optional[float] = Field(None, description = "Current loss value")
+    learning_rate: Optional[float] = Field(None, description = "Current learning rate")
     progress_percent: float = Field(
         ..., description = "Progress percentage (0.0 to 100.0)"
     )
@@ -196,3 +196,59 @@ class TrainingProgress(BaseModel):
     eval_loss: Optional[float] = Field(
         None, description = "Eval loss from the most recent evaluation step"
     )
+
+
+class TrainingRunSummary(BaseModel):
+    """Summary of a training run for list views."""
+
+    id: str
+    status: Literal["running", "completed", "stopped", "error"]
+    model_name: str
+    dataset_name: str
+    started_at: str
+    ended_at: Optional[str] = None
+    total_steps: Optional[int] = None
+    final_step: Optional[int] = None
+    final_loss: Optional[float] = None
+    output_dir: Optional[str] = None
+    duration_seconds: Optional[float] = None
+    error_message: Optional[str] = None
+    loss_sparkline: Optional[List[float]] = None
+
+
+class TrainingRunListResponse(BaseModel):
+    """Response for listing training runs."""
+
+    runs: List[TrainingRunSummary]
+    total: int
+
+
+class TrainingRunMetrics(BaseModel):
+    """Metrics arrays for a training run, using paired step arrays per metric."""
+
+    step_history: List[int] = Field(default_factory = list)
+    loss_history: List[float] = Field(default_factory = list)
+    loss_step_history: List[int] = Field(default_factory = list)
+    lr_history: List[float] = Field(default_factory = list)
+    lr_step_history: List[int] = Field(default_factory = list)
+    grad_norm_history: List[float] = Field(default_factory = list)
+    grad_norm_step_history: List[int] = Field(default_factory = list)
+    eval_loss_history: List[float] = Field(default_factory = list)
+    eval_step_history: List[int] = Field(default_factory = list)
+    final_epoch: Optional[float] = None
+    final_num_tokens: Optional[int] = None
+
+
+class TrainingRunDetailResponse(BaseModel):
+    """Response for a single training run with config and metrics."""
+
+    run: TrainingRunSummary
+    config: dict
+    metrics: TrainingRunMetrics
+
+
+class TrainingRunDeleteResponse(BaseModel):
+    """Response for deleting a training run."""
+
+    status: str
+    message: str
diff --git a/studio/backend/routes/__init__.py b/studio/backend/routes/__init__.py
index b45eff821b..e79f6553f9 100644
--- a/studio/backend/routes/__init__.py
+++ b/studio/backend/routes/__init__.py
@@ -12,6 +12,7 @@ from routes.datasets import router as datasets_router
 from routes.auth import router as auth_router
 from routes.data_recipe import router as data_recipe_router
 from routes.export import router as export_router
+from routes.training_history import router as training_history_router
 
 __all__ = [
     "training_router",
@@ -21,4 +22,5 @@ __all__ = [
     "auth_router",
     "data_recipe_router",
     "export_router",
+    "training_history_router",
 ]
diff --git a/studio/backend/routes/training.py b/studio/backend/routes/training.py
index 4f8054f80e..4cfb060dee 100644
--- a/studio/backend/routes/training.py
+++ b/studio/backend/routes/training.py
@@ -14,6 +14,7 @@ import structlog
 from loggers import get_logger
 import asyncio
 from datetime import datetime
+import uuid as _uuid
 
 # Add backend directory to path
 # The backend code should be in the same directory structure
@@ -115,15 +116,11 @@ async def start_training(
 
         backend = get_training_backend()
 
-        # Generate job ID and attach to backend for later status/progress calls
-        job_id = f"job_{datetime.now().strftime('%Y%m%d_%H%M%S')}"
-        backend.current_job_id = job_id
-
-        # Check if training is already active
+        # Check if training is already active (before mutating any state)
         if backend.is_training_active():
             existing_job_id: Optional[str] = getattr(backend, "current_job_id", "")
             return TrainingJobResponse(
-                job_id = existing_job_id or job_id,
+                job_id = existing_job_id or "",
                 status = "error",
                 message = (
                     "Training is already in progress. "
@@ -132,6 +129,12 @@ async def start_training(
                 error = "Training already active",
             )
 
+        # Generate job ID — passed into start_training() which sets it on the
+        # backend only after confirming the old pump thread is dead.
+        job_id = (
+            f"job_{datetime.now().strftime('%Y%m%d_%H%M%S')}_{_uuid.uuid4().hex[:8]}"
+        )
+
         # Validate dataset paths if provided
         if request.local_datasets:
             request.local_datasets = _validate_local_dataset_paths(
@@ -248,12 +251,12 @@ async def start_training(
             logger.warning("Could not shut down export subprocess: %s", e)
 
         # start_training now spawns a subprocess (non-blocking)
-        success = backend.start_training(**training_kwargs)
+        success = backend.start_training(job_id = job_id, **training_kwargs)
 
         if not success:
             progress_error = backend.trainer.training_progress.error
             return TrainingJobResponse(
-                job_id = job_id,
+                job_id = backend.current_job_id or "",
                 status = "error",
                 message = progress_error or "Failed to start training subprocess",
                 error = progress_error or "subprocess_start_failed",
@@ -345,7 +348,7 @@ async def reset_training(
             error = None,
             status_message = "Ready to train",
             step = 0,
-            loss = 0.0,
+            loss = None,
             epoch = 0,
             total_steps = 0,
         )
@@ -419,8 +422,8 @@ async def get_training_status(
                 "epoch": getattr(progress, "epoch", 0),
                 "step": getattr(progress, "step", 0),
                 "total_steps": getattr(progress, "total_steps", 0),
-                "loss": getattr(progress, "loss", 0.0),
-                "learning_rate": getattr(progress, "learning_rate", 0.0),
+                "loss": getattr(progress, "loss", None),
+                "learning_rate": getattr(progress, "learning_rate", None),
             }
 
         # Build metric history for chart recovery after SSE reconnection
@@ -526,8 +529,8 @@ async def stream_training_progress(
         # ── Helpers ──────────────────────────────────────────────
         def build_progress(
             step: int,
-            loss: float,
-            learning_rate: float,
+            loss: Optional[float],
+            learning_rate: Optional[float],
             total_steps: int,
             epoch: Optional[float] = None,
             progress: Optional[Any] = None,
@@ -604,10 +607,10 @@ async def stream_training_progress(
                     loss_val = (
                         backend.loss_history[i]
                         if i < len(backend.loss_history)
-                        else 0.0
+                        else None
                     )
                     lr_val = (
-                        backend.lr_history[i] if i < len(backend.lr_history) else 0.0
+                        backend.lr_history[i] if i < len(backend.lr_history) else None
                     )
                     tp_replay = getattr(
                         getattr(backend, "trainer", None), "training_progress", None
@@ -645,8 +648,8 @@ async def stream_training_progress(
 
             initial_progress = build_progress(
                 step = 0,
-                loss = 0.0,
-                learning_rate = 0.0,
+                loss = None,
+                learning_rate = None,
                 total_steps = initial_total_steps,
                 epoch = initial_epoch,
                 progress = tp,
@@ -660,9 +663,9 @@ async def stream_training_progress(
                 if backend.step_history:
                     final_step = backend.step_history[-1]
                     final_loss = (
-                        backend.loss_history[-1] if backend.loss_history else 0.0
+                        backend.loss_history[-1] if backend.loss_history else None
                     )
-                    final_lr = backend.lr_history[-1] if backend.lr_history else 0.0
+                    final_lr = backend.lr_history[-1] if backend.lr_history else None
                     final_total_steps = (
                         getattr(tp, "total_steps", final_step) if tp else final_step
                     )
@@ -680,7 +683,9 @@ async def stream_training_progress(
                     )
                 else:
                     yield format_sse(
-                        build_progress(-1, 0.0, 0.0, 0, progress = tp).model_dump_json(),
+                        build_progress(
+                            -1, None, None, 0, progress = tp
+                        ).model_dump_json(),
                         event = "complete",
                         event_id = 0,
                     )
@@ -698,9 +703,9 @@ async def stream_training_progress(
                 if backend.step_history:
                     current_step = backend.step_history[-1]
                     current_loss = (
-                        backend.loss_history[-1] if backend.loss_history else 0.0
+                        backend.loss_history[-1] if backend.loss_history else None
                     )
-                    current_lr = backend.lr_history[-1] if backend.lr_history else 0.0
+                    current_lr = backend.lr_history[-1] if backend.lr_history else None
                     tp_inner = getattr(
                         getattr(backend, "trainer", None), "training_progress", None
                     )
@@ -763,8 +768,8 @@ async def stream_training_progress(
                         )
                         preparing_payload = build_progress(
                             0,
-                            0.0,
-                            0.0,
+                            None,
+                            None,
                             prep_total,
                             progress = tp_prep,
                         )
@@ -781,7 +786,7 @@ async def stream_training_progress(
                         getattr(backend, "trainer", None), "training_progress", None
                     )
                     timeout_payload = build_progress(
-                        last_step, 0.0, 0.0, 0, progress = tp_timeout
+                        last_step, None, None, 0, progress = tp_timeout
                     )
                     yield format_sse(
                         timeout_payload.model_dump_json(),
@@ -797,7 +802,7 @@ async def stream_training_progress(
                 tp_error = getattr(
                     getattr(backend, "trainer", None), "training_progress", None
                 )
-                error_payload = build_progress(0, 0.0, 0.0, 0, progress = tp_error)
+                error_payload = build_progress(0, None, None, 0, progress = tp_error)
                 yield format_sse(
                     error_payload.model_dump_json(),
                     event = "error",
@@ -807,8 +812,8 @@ async def stream_training_progress(
 
         # ── Final "complete" event ───────────────────────────────
         final_step = backend.step_history[-1] if backend.step_history else last_step
-        final_loss = backend.loss_history[-1] if backend.loss_history else 0.0
-        final_lr = backend.lr_history[-1] if backend.lr_history else 0.0
+        final_loss = backend.loss_history[-1] if backend.loss_history else None
+        final_lr = backend.lr_history[-1] if backend.lr_history else None
         final_tp = getattr(getattr(backend, "trainer", None), "training_progress", None)
         final_total_steps = (
             getattr(final_tp, "total_steps", final_step) if final_tp else final_step
diff --git a/studio/backend/routes/training_history.py b/studio/backend/routes/training_history.py
new file mode 100644
index 0000000000..597c4424c0
--- /dev/null
+++ b/studio/backend/routes/training_history.py
@@ -0,0 +1,85 @@
+# SPDX-License-Identifier: AGPL-3.0-only
+# Copyright 2026-present the Unsloth AI Inc. team. All rights reserved. See /studio/LICENSE.AGPL-3.0
+
+"""
+Training history API routes — browse, view, and delete past training runs.
+"""
+
+import json
+
+from fastapi import APIRouter, Depends, HTTPException, Query
+from loggers import get_logger
+
+from auth.authentication import get_current_subject
+from models import (
+    TrainingRunDeleteResponse,
+    TrainingRunDetailResponse,
+    TrainingRunListResponse,
+    TrainingRunMetrics,
+    TrainingRunSummary,
+)
+from storage.studio_db import delete_run, get_run, get_run_metrics, list_runs
+
+logger = get_logger(__name__)
+
+router = APIRouter()
+
+
+@router.get("/runs", response_model = TrainingRunListResponse)
+async def list_training_runs(
+    limit: int = Query(50, ge = 1, le = 200),
+    offset: int = Query(0, ge = 0),
+    current_subject: str = Depends(get_current_subject),
+):
+    """List training runs, newest first."""
+    result = list_runs(limit = limit, offset = offset)
+    return TrainingRunListResponse(
+        runs = [TrainingRunSummary(**r) for r in result["runs"]],
+        total = result["total"],
+    )
+
+
+@router.get("/runs/{run_id}", response_model = TrainingRunDetailResponse)
+async def get_training_run_detail(
+    run_id: str,
+    current_subject: str = Depends(get_current_subject),
+):
+    """Get a single training run with full config and metrics."""
+    run = get_run(run_id)
+    if run is None:
+        raise HTTPException(status_code = 404, detail = f"Run {run_id} not found")
+
+    try:
+        config = json.loads(run.get("config_json", "{}"))
+    except (json.JSONDecodeError, TypeError):
+        logger.debug("Failed to parse config_json for run %s", run_id)
+        config = {}
+
+    metrics_data = get_run_metrics(run_id)
+
+    return TrainingRunDetailResponse(
+        run = TrainingRunSummary(**{k: v for k, v in run.items() if k != "config_json"}),
+        config = config,
+        metrics = TrainingRunMetrics(**metrics_data),
+    )
+
+
+@router.delete("/runs/{run_id}", response_model = TrainingRunDeleteResponse)
+async def delete_training_run(
+    run_id: str,
+    current_subject: str = Depends(get_current_subject),
+):
+    """Delete a training run and its metrics (CASCADE)."""
+    run = get_run(run_id)
+    if run is None:
+        raise HTTPException(status_code = 404, detail = f"Run {run_id} not found")
+    if run["status"] == "running":
+        raise HTTPException(
+            status_code = 409, detail = "Cannot delete a running training run"
+        )
+    logger.info("Deleting training run %s", run_id)
+    delete_run(run_id)
+    return TrainingRunDeleteResponse(
+        status = "deleted",
+        message = f"Run {run_id} deleted",
+    )
diff --git a/studio/backend/storage/__init__.py b/studio/backend/storage/__init__.py
new file mode 100644
index 0000000000..32014236c6
--- /dev/null
+++ b/studio/backend/storage/__init__.py
@@ -0,0 +1,2 @@
+# SPDX-License-Identifier: AGPL-3.0-only
+# Copyright 2026-present the Unsloth AI Inc. team. All rights reserved. See /studio/LICENSE.AGPL-3.0
diff --git a/studio/backend/storage/studio_db.py b/studio/backend/storage/studio_db.py
new file mode 100644
index 0000000000..4af19df42b
--- /dev/null
+++ b/studio/backend/storage/studio_db.py
@@ -0,0 +1,362 @@
+# SPDX-License-Identifier: AGPL-3.0-only
+# Copyright 2026-present the Unsloth AI Inc. team. All rights reserved. See /studio/LICENSE.AGPL-3.0
+
+"""
+SQLite storage for training run history and metrics.
+
+Follows the same pattern as auth/storage.py — module-level functions,
+raw sqlite3, per-function connections. Enhancements over auth:
+  - WAL mode for concurrent read/write access
+  - PRAGMA foreign_keys = ON for CASCADE deletes
+"""
+
+import json
+import logging
+import sqlite3
+import threading
+
+logger = logging.getLogger(__name__)
+from typing import Optional
+
+from utils.paths import studio_db_path, ensure_dir
+
+_schema_lock = threading.Lock()
+_schema_ready = False
+
+
+def _ensure_schema(conn: sqlite3.Connection) -> None:
+    """Create tables and indexes if they don't exist. Called once per process."""
+    conn.execute("PRAGMA journal_mode=WAL")
+    conn.execute(
+        """
+        CREATE TABLE IF NOT EXISTS training_runs (
+            id TEXT NOT NULL PRIMARY KEY,
+            status TEXT NOT NULL DEFAULT 'running',
+            model_name TEXT NOT NULL,
+            dataset_name TEXT NOT NULL,
+            config_json TEXT NOT NULL,
+            started_at TEXT NOT NULL,
+            ended_at TEXT,
+            total_steps INTEGER,
+            final_step INTEGER,
+            final_loss REAL,
+            output_dir TEXT,
+            error_message TEXT,
+            duration_seconds REAL,
+            loss_sparkline TEXT
+        )
+        """
+    )
+    conn.execute(
+        """
+        CREATE TABLE IF NOT EXISTS training_metrics (
+            id INTEGER PRIMARY KEY AUTOINCREMENT,
+            run_id TEXT NOT NULL REFERENCES training_runs(id) ON DELETE CASCADE,
+            step INTEGER NOT NULL,
+            loss REAL,
+            learning_rate REAL,
+            grad_norm REAL,
+            eval_loss REAL,
+            epoch REAL,
+            num_tokens INTEGER,
+            elapsed_seconds REAL,
+            UNIQUE(run_id, step)
+        )
+        """
+    )
+    conn.execute(
+        "CREATE INDEX IF NOT EXISTS idx_metrics_run_id ON training_metrics(run_id)"
+    )
+
+
+def get_connection() -> sqlite3.Connection:
+    """Open studio.db with WAL mode, create tables once per process, enable foreign keys."""
+    global _schema_ready
+    db_path = studio_db_path()
+    ensure_dir(db_path.parent)
+    conn = sqlite3.connect(str(db_path))
+    conn.row_factory = sqlite3.Row
+    # foreign_keys is session-scoped, must be set per connection
+    conn.execute("PRAGMA foreign_keys=ON")
+    if not _schema_ready:
+        with _schema_lock:
+            if not _schema_ready:
+                try:
+                    _ensure_schema(conn)
+                    _schema_ready = True
+                except Exception:
+                    conn.close()
+                    raise
+    return conn
+
+
+def create_run(
+    id: str,
+    model_name: str,
+    dataset_name: str,
+    config_json: str,
+    started_at: str,
+    total_steps: Optional[int],
+) -> None:
+    conn = get_connection()
+    try:
+        conn.execute(
+            """
+            INSERT INTO training_runs (id, model_name, dataset_name, config_json, started_at, total_steps)
+            VALUES (?, ?, ?, ?, ?, ?)
+            """,
+            (id, model_name, dataset_name, config_json, started_at, total_steps),
+        )
+        conn.commit()
+    finally:
+        conn.close()
+
+
+def update_run_total_steps(id: str, total_steps: int) -> None:
+    conn = get_connection()
+    try:
+        conn.execute(
+            "UPDATE training_runs SET total_steps = ? WHERE id = ?",
+            (total_steps, id),
+        )
+        conn.commit()
+    finally:
+        conn.close()
+
+
+def update_run_progress(
+    id: str, step: int, loss: Optional[float], duration_seconds: Optional[float]
+) -> None:
+    """Update current progress on a running training run (called on each metric flush)."""
+    conn = get_connection()
+    try:
+        conn.execute(
+            "UPDATE training_runs SET final_step = ?, final_loss = ?, duration_seconds = ? WHERE id = ?",
+            (step, loss, duration_seconds, id),
+        )
+        conn.commit()
+    finally:
+        conn.close()
+
+
+def finish_run(
+    id: str,
+    status: str,
+    ended_at: str,
+    final_step: Optional[int],
+    final_loss: Optional[float],
+    duration_seconds: Optional[float],
+    loss_sparkline: Optional[str] = None,
+    output_dir: Optional[str] = None,
+    error_message: Optional[str] = None,
+) -> None:
+    conn = get_connection()
+    try:
+        conn.execute(
+            """
+            UPDATE training_runs
+            SET status = ?, ended_at = ?, final_step = ?, final_loss = ?,
+                duration_seconds = ?, loss_sparkline = ?, output_dir = ?,
+                error_message = ?
+            WHERE id = ?
+            """,
+            (
+                status,
+                ended_at,
+                final_step,
+                final_loss,
+                duration_seconds,
+                loss_sparkline,
+                output_dir,
+                error_message,
+                id,
+            ),
+        )
+        conn.commit()
+    finally:
+        conn.close()
+
+
+def insert_metrics_batch(run_id: str, metrics: list[dict]) -> None:
+    if not metrics:
+        return
+    conn = get_connection()
+    try:
+        conn.executemany(
+            """
+            INSERT INTO training_metrics
+                (run_id, step, loss, learning_rate, grad_norm, eval_loss, epoch, num_tokens, elapsed_seconds)
+            VALUES (?, ?, ?, ?, ?, ?, ?, ?, ?)
+            ON CONFLICT(run_id, step) DO UPDATE SET
+                loss = COALESCE(excluded.loss, loss),
+                learning_rate = COALESCE(excluded.learning_rate, learning_rate),
+                grad_norm = COALESCE(excluded.grad_norm, grad_norm),
+                eval_loss = COALESCE(excluded.eval_loss, eval_loss),
+                epoch = COALESCE(excluded.epoch, epoch),
+                num_tokens = COALESCE(excluded.num_tokens, num_tokens),
+                elapsed_seconds = COALESCE(excluded.elapsed_seconds, elapsed_seconds)
+            """,
+            [
+                (
+                    run_id,
+                    m.get("step"),
+                    m.get("loss"),
+                    m.get("learning_rate"),
+                    m.get("grad_norm"),
+                    m.get("eval_loss"),
+                    m.get("epoch"),
+                    m.get("num_tokens"),
+                    m.get("elapsed_seconds"),
+                )
+                for m in metrics
+            ],
+        )
+        conn.commit()
+    finally:
+        conn.close()
+
+
+def list_runs(limit: int = 50, offset: int = 0) -> dict:
+    conn = get_connection()
+    try:
+        total = conn.execute("SELECT COUNT(*) FROM training_runs").fetchone()[0]
+        rows = conn.execute(
+            """
+            SELECT id, status, model_name, dataset_name, started_at, ended_at,
+                   total_steps, final_step, final_loss, output_dir,
+                   duration_seconds, error_message, loss_sparkline
+            FROM training_runs
+            ORDER BY started_at DESC
+            LIMIT ? OFFSET ?
+            """,
+            (limit, offset),
+        ).fetchall()
+        runs = []
+        for row in rows:
+            run = dict(row)
+            sparkline = run.get("loss_sparkline")
+            if sparkline:
+                try:
+                    run["loss_sparkline"] = json.loads(sparkline)
+                except (json.JSONDecodeError, TypeError):
+                    logger.debug(
+                        "Failed to parse loss_sparkline for run %s", run.get("id")
+                    )
+                    run["loss_sparkline"] = None
+            runs.append(run)
+        return {"runs": runs, "total": total}
+    finally:
+        conn.close()
+
+
+def get_run(id: str) -> Optional[dict]:
+    conn = get_connection()
+    try:
+        row = conn.execute("SELECT * FROM training_runs WHERE id = ?", (id,)).fetchone()
+        if row is None:
+            return None
+        run = dict(row)
+        sparkline = run.get("loss_sparkline")
+        if sparkline:
+            try:
+                run["loss_sparkline"] = json.loads(sparkline)
+            except (json.JSONDecodeError, TypeError):
+                logger.debug("Failed to parse loss_sparkline for run %s", id)
+                run["loss_sparkline"] = None
+        return run
+    finally:
+        conn.close()
+
+
+def get_run_metrics(id: str) -> dict:
+    """Return metric arrays for a run, using paired step arrays per metric."""
+    conn = get_connection()
+    try:
+        rows = conn.execute(
+            """
+            SELECT step, loss, learning_rate, grad_norm, eval_loss, epoch,
+                   num_tokens, elapsed_seconds
+            FROM training_metrics
+            WHERE run_id = ?
+            ORDER BY step
+            """,
+            (id,),
+        ).fetchall()
+
+        step_history: list[int] = []
+        loss_history: list[float] = []
+        loss_step_history: list[int] = []
+        lr_history: list[float] = []
+        lr_step_history: list[int] = []
+        grad_norm_history: list[float] = []
+        grad_norm_step_history: list[int] = []
+        eval_loss_history: list[float] = []
+        eval_step_history: list[int] = []
+        final_epoch: float | None = None
+        final_num_tokens: int | None = None
+
+        for row in rows:
+            step = row["step"]
+            step_history.append(step)
+            if step > 0 and row["loss"] is not None:
+                loss_history.append(row["loss"])
+                loss_step_history.append(step)
+            if step > 0 and row["learning_rate"] is not None:
+                lr_history.append(row["learning_rate"])
+                lr_step_history.append(step)
+            if step > 0 and row["grad_norm"] is not None:
+                grad_norm_history.append(row["grad_norm"])
+                grad_norm_step_history.append(step)
+            if step > 0 and row["eval_loss"] is not None:
+                eval_loss_history.append(row["eval_loss"])
+                eval_step_history.append(step)
+            if row["epoch"] is not None:
+                final_epoch = row["epoch"]
+            if row["num_tokens"] is not None:
+                final_num_tokens = row["num_tokens"]
+
+        return {
+            "step_history": step_history,
+            "loss_history": loss_history,
+            "loss_step_history": loss_step_history,
+            "lr_history": lr_history,
+            "lr_step_history": lr_step_history,
+            "grad_norm_history": grad_norm_history,
+            "grad_norm_step_history": grad_norm_step_history,
+            "eval_loss_history": eval_loss_history,
+            "eval_step_history": eval_step_history,
+            "final_epoch": final_epoch,
+            "final_num_tokens": final_num_tokens,
+        }
+    finally:
+        conn.close()
+
+
+def delete_run(id: str) -> None:
+    conn = get_connection()
+    try:
+        conn.execute("DELETE FROM training_runs WHERE id = ?", (id,))
+        conn.commit()
+    finally:
+        conn.close()
+
+
+def cleanup_orphaned_runs() -> None:
+    """Mark any 'running' rows as errored on startup (server restarted mid-training)."""
+    from datetime import datetime, timezone
+
+    conn = get_connection()
+    try:
+        conn.execute(
+            """
+            UPDATE training_runs
+            SET status = 'error',
+                error_message = 'Server restarted during training',
+                ended_at = ?
+            WHERE status = 'running'
+            """,
+            (datetime.now(timezone.utc).isoformat(),),
+        )
+        conn.commit()
+    finally:
+        conn.close()
diff --git a/studio/backend/utils/downsample.py b/studio/backend/utils/downsample.py
new file mode 100644
index 0000000000..bccf6a23b7
--- /dev/null
+++ b/studio/backend/utils/downsample.py
@@ -0,0 +1,18 @@
+# SPDX-License-Identifier: AGPL-3.0-only
+# Copyright 2026-present the Unsloth AI Inc. team. All rights reserved. See /studio/LICENSE.AGPL-3.0
+
+"""Generic numeric downsampling utility."""
+
+
+def downsample(values: list[float], target_count: int) -> list[float]:
+    """Reduce a list to target_count points via evenly-spaced index sampling."""
+    if len(values) <= target_count:
+        return list(values)
+    if target_count <= 0:
+        return []
+    if target_count == 1:
+        return [values[-1]]
+    indices = [
+        round(i * (len(values) - 1) / (target_count - 1)) for i in range(target_count)
+    ]
+    return [values[i] for i in indices]
diff --git a/studio/backend/utils/paths/__init__.py b/studio/backend/utils/paths/__init__.py
index 90df216f96..789052f372 100644
--- a/studio/backend/utils/paths/__init__.py
+++ b/studio/backend/utils/paths/__init__.py
@@ -16,6 +16,7 @@ from .storage_roots import (
     exports_root,
     auth_root,
     auth_db_path,
+    studio_db_path,
     tmp_root,
     seed_uploads_root,
     unstructured_seed_cache_root,
@@ -45,6 +46,7 @@ __all__ = [
     "exports_root",
     "auth_root",
     "auth_db_path",
+    "studio_db_path",
     "tmp_root",
     "seed_uploads_root",
     "unstructured_seed_cache_root",
diff --git a/studio/backend/utils/paths/storage_roots.py b/studio/backend/utils/paths/storage_roots.py
index f14887e119..626e868275 100644
--- a/studio/backend/utils/paths/storage_roots.py
+++ b/studio/backend/utils/paths/storage_roots.py
@@ -49,6 +49,10 @@ def auth_db_path() -> Path:
     return auth_root() / "auth.db"
 
 
+def studio_db_path() -> Path:
+    return studio_root() / "studio.db"
+
+
 def tmp_root() -> Path:
     return Path(tempfile.gettempdir()) / "unsloth-studio"
 
diff --git a/studio/frontend/src/features/studio/historical-training-view.tsx b/studio/frontend/src/features/studio/historical-training-view.tsx
new file mode 100644
index 0000000000..d3ac434bd8
--- /dev/null
+++ b/studio/frontend/src/features/studio/historical-training-view.tsx
@@ -0,0 +1,168 @@
+// SPDX-License-Identifier: AGPL-3.0-only
+// Copyright 2026-present the Unsloth AI Inc. team. All rights reserved. See /studio/LICENSE.AGPL-3.0
+
+import type { TrainingViewData } from "@/features/training";
+import { getTrainingRun } from "@/features/training";
+import type { TrainingRunDetailResponse } from "@/features/training";
+import { type ReactElement, useEffect, useState } from "react";
+import { ChartsSection } from "./sections/charts-section";
+import { ProgressSection } from "./sections/progress-section";
+
+interface HistoricalTrainingViewProps {
+  runId: string;
+}
+
+function normalizeTrainingMethod(config: Record<string, unknown>): string {
+  const type = config?.training_type as string | undefined;
+  if (!type || type === "Full Finetuning") return "full";
+  if (type === "LoRA/QLoRA") {
+    return config?.load_in_4bit ? "qlora" : "lora";
+  }
+  return "full";
+}
+
+function mapToViewData(detail: TrainingRunDetailResponse): TrainingViewData {
+  const { run, metrics } = detail;
+
+  const lossHistory = metrics.loss_step_history
+    .map((step, i) => ({ step, value: metrics.loss_history[i] }))
+    .filter((p): p is { step: number; value: number } => p.value != null);
+
+  const lrHistory = metrics.lr_step_history
+    .map((step, i) => ({ step, value: metrics.lr_history[i] }))
+    .filter((p): p is { step: number; value: number } => p.value != null);
+
+  const gradNormHistory = metrics.grad_norm_step_history
+    .map((step, i) => ({ step, value: metrics.grad_norm_history[i] }))
+    .filter((p): p is { step: number; value: number } => p.value != null);
+
+  const evalLossHistory = metrics.eval_step_history
+    .map((step, i) => ({ step, value: metrics.eval_loss_history[i] }))
+    .filter((p): p is { step: number; value: number } => p.value != null);
+
+  const phase =
+    run.status === "completed"
+      ? "completed"
+      : run.status === "stopped"
+        ? "stopped"
+        : run.status === "error"
+          ? "error"
+          : run.status === "running"
+            ? "training"
+            : "idle";
+
+  return {
+    phase,
+    currentStep: run.final_step ?? 0,
+    totalSteps: run.total_steps ?? 0,
+    currentLoss: run.final_loss,
+    currentLearningRate: metrics.lr_history.at(-1) ?? null,
+    currentGradNorm: metrics.grad_norm_history.at(-1) ?? null,
+    currentEpoch: metrics.final_epoch,
+    currentNumTokens: metrics.final_num_tokens ?? null,
+    progressPercent:
+      run.total_steps && run.final_step
+        ? (run.final_step / run.total_steps) * 100
+        : 0,
+    elapsedSeconds: run.duration_seconds,
+    etaSeconds: null,
+    evalEnabled: evalLossHistory.length > 0,
+    message:
+      run.status === "completed"
+        ? "Training completed"
+        : run.status === "stopped"
+          ? "Training stopped"
+          : run.status === "running"
+            ? "Training in progress"
+            : run.error_message ?? "Training errored",
+    error: run.status === "error" ? run.error_message : null,
+    isTrainingRunning: false,
+    modelName: run.model_name,
+    trainingMethod: normalizeTrainingMethod(detail.config),
+    lossHistory,
+    lrHistory,
+    gradNormHistory,
+    evalLossHistory,
+  };
+}
+
+export function HistoricalTrainingView({
+  runId,
+}: HistoricalTrainingViewProps): ReactElement {
+  const [detail, setDetail] = useState<TrainingRunDetailResponse | null>(null);
+  const [error, setError] = useState<string | null>(null);
+
+  // Derive loading from detail/error -- no separate state needed
+  const loading = detail === null && error === null;
+
+  useEffect(() => {
+    const controller = new AbortController();
+    getTrainingRun(runId, controller.signal)
+      .then((result) => {
+        setDetail(result);
+      })
+      .catch((err) => {
+        if (err instanceof DOMException && err.name === "AbortError") return;
+        setError(err instanceof Error ? err.message : "Failed to load run");
+      });
+    return () => {
+      controller.abort();
+      // Reset on runId change so loading derives correctly for the next fetch
+      setDetail(null);
+      setError(null);
+    };
+  }, [runId]);
+
+  if (loading) {
+    return (
+      <div className="rounded-xl border bg-card p-8 text-sm text-muted-foreground">
+        Loading training run...
+      </div>
+    );
+  }
+
+  if (error || !detail) {
+    return (
+      <div className="rounded-xl border border-destructive/30 bg-destructive/5 p-8 text-sm text-red-500">
+        {error ?? "Run not found"}
+      </div>
+    );
+  }
+
+  const viewData = mapToViewData(detail);
+  const configOverride = detail.config
+    ? {
+        epochs: detail.config.num_epochs as number | undefined,
+        batchSize: detail.config.batch_size as number | undefined,
+        learningRate: detail.config.learning_rate as string | undefined,
+        maxSteps: detail.config.max_steps as number | undefined,
+        contextLength: detail.config.max_seq_length as number | undefined,
+        warmupSteps: detail.config.warmup_steps as number | undefined,
+        optimizerType: detail.config.optim as string | undefined,
+        loraRank: detail.config.lora_r as number | undefined,
+        loraAlpha: detail.config.lora_alpha as number | undefined,
+        loraDropout: detail.config.lora_dropout as number | undefined,
+        loraVariant: detail.config.use_rslora ? "rsLoRA" : undefined,
+      }
+    : undefined;
+
+  return (
+    <div className="flex flex-col gap-6">
+      <ProgressSection
+        data={viewData}
+        isHistorical
+        configOverride={configOverride}
+      />
+      <ChartsSection
+        currentStep={viewData.currentStep}
+        totalSteps={viewData.totalSteps}
+        isTraining={false}
+        evalEnabled={viewData.evalEnabled}
+        lossHistory={viewData.lossHistory}
+        lrHistory={viewData.lrHistory}
+        gradNormHistory={viewData.gradNormHistory}
+        evalLossHistory={viewData.evalLossHistory}
+      />
+    </div>
+  );
+}
diff --git a/studio/frontend/src/features/studio/history-card-grid.tsx b/studio/frontend/src/features/studio/history-card-grid.tsx
new file mode 100644
index 0000000000..78859d2f81
--- /dev/null
+++ b/studio/frontend/src/features/studio/history-card-grid.tsx
@@ -0,0 +1,401 @@
+// SPDX-License-Identifier: AGPL-3.0-only
+// Copyright 2026-present the Unsloth AI Inc. team. All rights reserved. See /studio/LICENSE.AGPL-3.0
+
+import {
+  AlertDialog,
+  AlertDialogAction,
+  AlertDialogCancel,
+  AlertDialogContent,
+  AlertDialogDescription,
+  AlertDialogFooter,
+  AlertDialogHeader,
+  AlertDialogTitle,
+} from "@/components/ui/alert-dialog";
+import { Button } from "@/components/ui/button";
+import type { TrainingRunSummary } from "@/features/training";
+import { deleteTrainingRun, listTrainingRuns } from "@/features/training";
+import { cn } from "@/lib/utils";
+import { Delete02Icon } from "@hugeicons/core-free-icons";
+import { HugeiconsIcon } from "@hugeicons/react";
+import { type ReactElement, useCallback, useEffect, useRef, useState } from "react";
+import { Spinner } from "@/components/ui/spinner";
+
+const PAGE_SIZE = 12;
+const RUNNING_POLL_INTERVAL_MS = 5000;
+
+const statusBadge: Record<
+  string,
+  { label: string; className: string }
+> = {
+  completed: {
+    label: "Completed",
+    className:
+      "bg-emerald-100 text-emerald-700 dark:bg-emerald-950 dark:text-emerald-400",
+  },
+  stopped: {
+    label: "Stopped",
+    className:
+      "bg-amber-100 text-amber-700 dark:bg-amber-950 dark:text-amber-400",
+  },
+  error: {
+    label: "Error",
+    className: "bg-red-100 text-red-700 dark:bg-red-950 dark:text-red-400",
+  },
+  running: {
+    label: "Running",
+    className:
+      "bg-blue-100 text-blue-700 dark:bg-blue-950 dark:text-blue-400",
+  },
+};
+
+function catmullRomPath(points: { x: number; y: number }[]): string {
+  if (points.length < 2) return "";
+  const d = [`M${points[0]!.x.toFixed(1)},${points[0]!.y.toFixed(1)}`];
+  for (let i = 0; i < points.length - 1; i++) {
+    const p0 = points[Math.max(i - 1, 0)]!;
+    const p1 = points[i]!;
+    const p2 = points[i + 1]!;
+    const p3 = points[Math.min(i + 2, points.length - 1)]!;
+    const cp1x = p1.x + (p2.x - p0.x) / 6;
+    const cp1y = p1.y + (p2.y - p0.y) / 6;
+    const cp2x = p2.x - (p3.x - p1.x) / 6;
+    const cp2y = p2.y - (p3.y - p1.y) / 6;
+    d.push(
+      `C${cp1x.toFixed(1)},${cp1y.toFixed(1)} ${cp2x.toFixed(1)},${cp2y.toFixed(1)} ${p2.x.toFixed(1)},${p2.y.toFixed(1)}`,
+    );
+  }
+  return d.join(" ");
+}
+
+function Sparkline({ values, id }: { values: number[]; id: string }): ReactElement | null {
+  if (!values || values.length < 2) return null;
+  let min = values[0]!;
+  let max = values[0]!;
+  for (let i = 1; i < values.length; i++) {
+    if (values[i]! < min) min = values[i]!;
+    if (values[i]! > max) max = values[i]!;
+  }
+  const range = max - min || 1;
+  const pad = 1.5; // half stroke-width so peaks aren't clipped
+  const h = 32;
+  const w = 120;
+  const gradientId = `sparkFill-${id}`;
+
+  // Build points with vertical padding so the stroke isn't clipped
+  const pts = values.map((v, i) => ({
+    x: (i / (values.length - 1)) * w,
+    y: pad + (1 - (v - min) / range) * (h - pad * 2),
+  }));
+
+  const linePath = catmullRomPath(pts);
+  const last = pts[pts.length - 1]!;
+  const first = pts[0]!;
+  const fillPath = `${linePath} L${last.x.toFixed(1)},${h} L${first.x.toFixed(1)},${h} Z`;
+
+  return (
+    <svg viewBox={`0 0 ${w} ${h}`} className="h-8 w-full" preserveAspectRatio="none" role="img" aria-label="Loss trend sparkline">
+      <defs>
+        <linearGradient id={gradientId} x1="0" y1="0" x2="0" y2="1">
+          <stop offset="0%" stopColor="currentColor" stopOpacity="0.12" />
+          <stop offset="100%" stopColor="currentColor" stopOpacity="0" />
+        </linearGradient>
+      </defs>
+      <path
+        d={fillPath}
+        fill={`url(#${gradientId})`}
+        className="text-emerald-500"
+      />
+      <path
+        d={linePath}
+        fill="none"
+        stroke="currentColor"
+        strokeWidth="1.5"
+        strokeLinecap="round"
+        strokeLinejoin="round"
+        className="text-emerald-500"
+      />
+    </svg>
+  );
+}
+
+function formatRelativeTime(isoDate: string): string {
+  const diff = Date.now() - new Date(isoDate).getTime();
+  const mins = Math.floor(diff / 60000);
+  if (mins < 1) return "just now";
+  if (mins < 60) return `${mins}m ago`;
+  const hrs = Math.floor(mins / 60);
+  if (hrs < 24) return `${hrs}h ago`;
+  const days = Math.floor(hrs / 24);
+  return `${days}d ago`;
+}
+
+function formatDuration(seconds: number | null): string {
+  if (seconds == null) return "--";
+  const total = Math.floor(seconds);
+  if (total < 60) return `${total}s`;
+  const min = Math.floor(total / 60);
+  const sec = total % 60;
+  if (min < 60) return `${min}m ${sec}s`;
+  const hrs = Math.floor(min / 60);
+  return `${hrs}h ${min % 60}m`;
+}
+
+interface HistoryCardGridProps {
+  onSelectRun: (runId: string) => void;
+}
+
+export function HistoryCardGrid({
+  onSelectRun,
+}: HistoryCardGridProps): ReactElement {
+  const [runs, setRuns] = useState<TrainingRunSummary[]>([]);
+  const [total, setTotal] = useState(0);
+  const [loading, setLoading] = useState(true);
+  const [error, setError] = useState<string | null>(null);
+  const [deleteError, setDeleteError] = useState<string | null>(null);
+  const [deleteTarget, setDeleteTarget] = useState<string | null>(null);
+  const [manualFetchInFlight, setManualFetchInFlight] = useState(false);
+
+  const userControllerRef = useRef<AbortController | null>(null);
+  const pollControllerRef = useRef<AbortController | null>(null);
+  const fetchIdRef = useRef(0);
+  const pollIdRef = useRef(0);
+
+  const fetchRuns = useCallback(async (offset = 0, append = false, limit = PAGE_SIZE) => {
+    // Cancel any in-flight poll so its stale response can't clobber this fresher fetch
+    pollControllerRef.current?.abort();
+    userControllerRef.current?.abort();
+    const controller = new AbortController();
+    userControllerRef.current = controller;
+    const id = ++fetchIdRef.current;
+
+    setManualFetchInFlight(true);
+    setLoading(true);
+    setError(null);
+    try {
+      const result = await listTrainingRuns(limit, offset, controller.signal);
+      if (fetchIdRef.current !== id) return;
+      setRuns((prev) => (append ? [...prev, ...result.runs] : result.runs));
+      setTotal(result.total);
+    } catch (err) {
+      if (err instanceof DOMException && err.name === "AbortError") return;
+      if (fetchIdRef.current !== id) return;
+      if (!append) setError("Failed to load training runs");
+    } finally {
+      if (fetchIdRef.current === id) {
+        setLoading(false);
+        setManualFetchInFlight(false);
+      }
+    }
+  }, []);
+
+  useEffect(() => {
+    void fetchRuns(0);
+    return () => {
+      userControllerRef.current?.abort();
+    };
+  }, [fetchRuns]);
+
+  // Poll while any run is still "running" so the card shows live progress
+  const hasRunningRun = runs.some((r) => r.status === "running");
+  const visibleCount = runs.length;
+  useEffect(() => {
+    if (!hasRunningRun) return;
+    const timer = setInterval(async () => {
+      if (manualFetchInFlight) return;
+      pollControllerRef.current?.abort();
+      const controller = new AbortController();
+      pollControllerRef.current = controller;
+      const pid = ++pollIdRef.current;
+      try {
+        const limit = Math.max(PAGE_SIZE, visibleCount);
+        const result = await listTrainingRuns(limit, 0, controller.signal);
+        if (pollIdRef.current !== pid) return; // stale poll — discard
+        setRuns(result.runs);
+        setTotal(result.total);
+      } catch {
+        // silently handle — poll will retry
+      }
+    }, RUNNING_POLL_INTERVAL_MS);
+    return () => {
+      clearInterval(timer);
+      pollControllerRef.current?.abort();
+    };
+  }, [hasRunningRun, visibleCount, manualFetchInFlight]);
+
+  const handleDelete = async () => {
+    if (!deleteTarget) return;
+    setDeleteError(null);
+    try {
+      await deleteTrainingRun(deleteTarget);
+      // Optimistically remove the card so it disappears immediately
+      setRuns((prev) => prev.filter((r) => r.id !== deleteTarget));
+      setTotal((prev) => Math.max(0, prev - 1));
+      // Re-fetch preserving visible count so offsets stay consistent for "Load more"
+      const currentCount = runs.length - 1;
+      const limit = Math.max(PAGE_SIZE, currentCount);
+      fetchRuns(0, false, limit).catch(() => {
+        // Refresh failed — card is already removed, no stale display
+      });
+    } catch {
+      setDeleteError("Failed to delete training run. Please try again.");
+    }
+    setDeleteTarget(null);
+  };
+
+  if (!loading && error && runs.length === 0) {
+    return (
+      <div className="flex flex-col items-center gap-2 py-16 text-center">
+        <p className="text-sm text-destructive">{error}</p>
+        <Button variant="outline" size="sm" onClick={() => void fetchRuns(0)}>
+          Retry
+        </Button>
+      </div>
+    );
+  }
+
+  if (!loading && runs.length === 0) {
+    return (
+      <div className="flex flex-col items-center gap-2 py-16 text-center">
+        <p className="text-sm text-muted-foreground">
+          No training runs yet. Start your first training run in the Configure
+          tab.
+        </p>
+      </div>
+    );
+  }
+
+  return (
+    <>
+      {deleteError && (
+        <div className="mb-4 rounded-lg border border-destructive/50 bg-destructive/10 px-4 py-2 text-sm text-destructive">
+          {deleteError}
+        </div>
+      )}
+      <div className="grid grid-cols-1 gap-4 sm:grid-cols-2 xl:grid-cols-3">
+        {runs.map((run) => {
+          const badge = statusBadge[run.status] ?? statusBadge.error;
+          const isRunning = run.status === "running";
+          return (
+            <div
+              role="button"
+              tabIndex={0}
+              key={run.id}
+              className={cn(
+                "group relative flex cursor-pointer flex-col gap-3 rounded-xl border bg-card p-4 text-left transition-colors hover:border-border hover:bg-accent/30",
+                isRunning
+                  ? "border-blue-400/50 dark:border-blue-500/30"
+                  : "border-border/60",
+              )}
+              onClick={() => onSelectRun(run.id)}
+              onKeyDown={(e) => {
+                if (e.key === "Enter" || e.key === " ") {
+                  e.preventDefault();
+                  onSelectRun(run.id);
+                }
+              }}
+            >
+              <div className="flex items-center justify-between pr-6">
+                <span
+                  className={cn(
+                    "inline-flex items-center gap-1.5 rounded-full px-2 py-0.5 text-[10px] font-semibold",
+                    badge.className,
+                  )}
+                >
+                  {isRunning && <Spinner className="size-2.5" />}
+                  {badge.label}
+                </span>
+                <span className="text-[10px] text-muted-foreground">
+                  {formatRelativeTime(run.started_at)}
+                </span>
+              </div>
+              <div className="min-w-0">
+                <p
+                  className="truncate text-sm font-medium"
+                  title={run.model_name}
+                >
+                  {run.model_name}
+                </p>
+                <p className="truncate text-xs text-muted-foreground">
+                  {run.dataset_name}
+                </p>
+              </div>
+              {run.loss_sparkline && run.loss_sparkline.length >= 2 && (
+                <Sparkline values={run.loss_sparkline} id={run.id} />
+              )}
+              <div className="flex flex-wrap gap-x-4 gap-y-1 text-[11px] text-muted-foreground">
+                <span>
+                  Loss:{" "}
+                  {run.final_loss != null ? run.final_loss.toFixed(4) : "--"}
+                </span>
+                <span>
+                  Steps: {run.final_step ?? 0}/{run.total_steps ?? "--"}
+                </span>
+                <span>{formatDuration(run.duration_seconds)}</span>
+              </div>
+              {!isRunning && (
+                <button
+                  type="button"
+                  className="absolute right-3 top-3 rounded-md p-1 text-muted-foreground/50 opacity-0 transition-opacity hover:bg-destructive/10 hover:text-destructive group-hover:opacity-100 focus-visible:opacity-100"
+                  aria-label="Delete run"
+                  onClick={(e) => {
+                    e.stopPropagation();
+                    setDeleteTarget(run.id);
+                  }}
+                >
+                  <HugeiconsIcon icon={Delete02Icon} className="size-3.5" />
+                </button>
+              )}
+            </div>
+          );
+        })}
+      </div>
+      {runs.length < total && (
+        <div className="mt-4 flex justify-center">
+          <Button
+            variant="outline"
+            size="sm"
+            onClick={() => void fetchRuns(runs.length, true)}
+            disabled={loading}
+          >
+            {loading ? "Loading..." : "Load more"}
+          </Button>
+        </div>
+      )}
+      {loading && runs.length === 0 && (
+        <div className="grid grid-cols-1 gap-4 sm:grid-cols-2 xl:grid-cols-3">
+          {Array.from({ length: 3 }).map((_, i) => (
+            <div
+              key={`skeleton-${i}`}
+              className="h-40 animate-pulse rounded-xl border bg-muted/30"
+            />
+          ))}
+        </div>
+      )}
+      <AlertDialog
+        open={deleteTarget !== null}
+        onOpenChange={(open) => {
+          if (!open) setDeleteTarget(null);
+        }}
+      >
+        <AlertDialogContent>
+          <AlertDialogHeader>
+            <AlertDialogTitle>Delete training run?</AlertDialogTitle>
+            <AlertDialogDescription>
+              This will permanently delete this training run and all its metrics.
+              This action cannot be undone.
+            </AlertDialogDescription>
+          </AlertDialogHeader>
+          <AlertDialogFooter>
+            <AlertDialogCancel>Cancel</AlertDialogCancel>
+            <AlertDialogAction
+              variant="destructive"
+              onClick={() => void handleDelete()}
+            >
+              Delete
+            </AlertDialogAction>
+          </AlertDialogFooter>
+        </AlertDialogContent>
+      </AlertDialog>
+    </>
+  );
+}
diff --git a/studio/frontend/src/features/studio/live-training-view.tsx b/studio/frontend/src/features/studio/live-training-view.tsx
new file mode 100644
index 0000000000..9f930ce77b
--- /dev/null
+++ b/studio/frontend/src/features/studio/live-training-view.tsx
@@ -0,0 +1,118 @@
+// SPDX-License-Identifier: AGPL-3.0-only
+// Copyright 2026-present the Unsloth AI Inc. team. All rights reserved. See /studio/LICENSE.AGPL-3.0
+
+import { cn } from "@/lib/utils";
+import {
+  useTrainingConfigStore,
+  useTrainingRuntimeStore,
+} from "@/features/training";
+import type { TrainingViewData } from "@/features/training";
+import type { ReactElement } from "react";
+import { useShallow } from "zustand/react/shallow";
+import { ChartsSection } from "./sections/charts-section";
+import { ProgressSection } from "./sections/progress-section";
+import { TrainingStartOverlay } from "./training-start-overlay";
+
+export function LiveTrainingView(): ReactElement {
+  const runtime = useTrainingRuntimeStore(
+    useShallow((state) => ({
+      jobId: state.jobId,
+      phase: state.phase,
+      message: state.message,
+      error: state.error,
+      currentStep: state.currentStep,
+      totalSteps: state.totalSteps,
+      currentEpoch: state.currentEpoch,
+      currentLoss: state.currentLoss,
+      currentLearningRate: state.currentLearningRate,
+      currentGradNorm: state.currentGradNorm,
+      currentNumTokens: state.currentNumTokens,
+      progressPercent: state.progressPercent,
+      elapsedSeconds: state.elapsedSeconds,
+      etaSeconds: state.etaSeconds,
+      evalEnabled: state.evalEnabled,
+      isTrainingRunning: state.isTrainingRunning,
+      lossHistory: state.lossHistory,
+      lrHistory: state.lrHistory,
+      gradNormHistory: state.gradNormHistory,
+      evalLossHistory: state.evalLossHistory,
+      firstStepReceived: state.firstStepReceived,
+      isStarting: state.isStarting,
+    })),
+  );
+
+  const config = useTrainingConfigStore(
+    useShallow((state) => ({
+      selectedModel: state.selectedModel,
+      trainingMethod: state.trainingMethod,
+    })),
+  );
+
+  const viewData: TrainingViewData = {
+    phase: runtime.phase,
+    currentStep: runtime.currentStep,
+    totalSteps: runtime.totalSteps,
+    currentLoss: runtime.currentLoss,
+    currentLearningRate: runtime.currentLearningRate,
+    currentGradNorm: runtime.currentGradNorm,
+    currentEpoch: runtime.currentEpoch,
+    currentNumTokens: runtime.currentNumTokens,
+    progressPercent: runtime.progressPercent,
+    elapsedSeconds: runtime.elapsedSeconds,
+    etaSeconds: runtime.etaSeconds,
+    evalEnabled: runtime.evalEnabled,
+    message: runtime.message,
+    error: runtime.error,
+    isTrainingRunning: runtime.isTrainingRunning,
+    modelName: config.selectedModel ?? "",
+    trainingMethod: config.trainingMethod ?? "",
+    lossHistory: runtime.lossHistory,
+    lrHistory: runtime.lrHistory,
+    gradNormHistory: runtime.gradNormHistory,
+    evalLossHistory: runtime.evalLossHistory,
+  };
+
+  const isPreparingPhase =
+    runtime.phase === "downloading_model" ||
+    runtime.phase === "downloading_dataset" ||
+    runtime.phase === "loading_model" ||
+    runtime.phase === "loading_dataset" ||
+    runtime.phase === "configuring";
+  const isWaitingForFirstStep =
+    runtime.phase === "training" && !runtime.firstStepReceived;
+  const showOverlay =
+    runtime.isStarting ||
+    isPreparingPhase ||
+    (isWaitingForFirstStep && runtime.currentStep <= 0);
+
+  return (
+    <div className={cn("relative", showOverlay && "min-h-[72vh]")}>
+      <div
+        className={cn(
+          "relative z-10 flex flex-col gap-6 transition-[filter]",
+          showOverlay && "blur",
+        )}
+      >
+        <div data-tour="studio-training-progress">
+          <ProgressSection key={runtime.jobId ?? "no-job"} data={viewData} />
+        </div>
+        <ChartsSection
+          currentStep={viewData.currentStep}
+          totalSteps={viewData.totalSteps}
+          isTraining={viewData.isTrainingRunning}
+          evalEnabled={viewData.evalEnabled}
+          lossHistory={viewData.lossHistory}
+          lrHistory={viewData.lrHistory}
+          gradNormHistory={viewData.gradNormHistory}
+          evalLossHistory={viewData.evalLossHistory}
+        />
+      </div>
+      {showOverlay ? (
+        <TrainingStartOverlay
+          message={runtime.message}
+          currentStep={runtime.currentStep}
+        />
+      ) : null}
+    </div>
+  );
+}
diff --git a/studio/frontend/src/features/studio/sections/charts-section.tsx b/studio/frontend/src/features/studio/sections/charts-section.tsx
index a7dacf7e5b..4c7bf46d9d 100644
--- a/studio/frontend/src/features/studio/sections/charts-section.tsx
+++ b/studio/frontend/src/features/studio/sections/charts-section.tsx
@@ -1,7 +1,7 @@
 // SPDX-License-Identifier: AGPL-3.0-only
 // Copyright 2026-present the Unsloth AI Inc. team. All rights reserved. See /studio/LICENSE.AGPL-3.0
 
-import { useTrainingRuntimeStore } from "@/features/training";
+import type { TrainingSeriesPoint } from "@/features/training";
 import { type ReactElement, Suspense, lazy, useMemo } from "react";
 
 const ChartsContent = lazy(() =>
@@ -16,42 +16,49 @@ const SKELETON_KEYS = [
   "chart-skeleton-4",
 ];
 
-export function ChartsSection(): ReactElement | null {
-  const currentStep = useTrainingRuntimeStore((state) => state.currentStep);
-  const totalSteps = useTrainingRuntimeStore((state) => state.totalSteps);
-  const isTraining = useTrainingRuntimeStore((state) => state.isTrainingRunning);
-  const evalEnabled = useTrainingRuntimeStore((state) => state.evalEnabled);
-  const lossHistoryRaw = useTrainingRuntimeStore((state) => state.lossHistory);
-  const lrHistoryRaw = useTrainingRuntimeStore((state) => state.lrHistory);
-  const gradNormHistoryRaw = useTrainingRuntimeStore(
-    (state) => state.gradNormHistory,
-  );
-  const evalLossHistoryRaw = useTrainingRuntimeStore(
-    (state) => state.evalLossHistory,
-  );
+interface ChartsSectionProps {
+  currentStep: number;
+  totalSteps: number;
+  isTraining: boolean;
+  evalEnabled: boolean;
+  lossHistory: TrainingSeriesPoint[];
+  lrHistory: TrainingSeriesPoint[];
+  gradNormHistory: TrainingSeriesPoint[];
+  evalLossHistory: TrainingSeriesPoint[];
+}
 
+export function ChartsSection({
+  currentStep,
+  totalSteps,
+  isTraining,
+  evalEnabled,
+  lossHistory,
+  lrHistory,
+  gradNormHistory,
+  evalLossHistory,
+}: ChartsSectionProps): ReactElement | null {
   const series = useMemo(
     () => ({
       currentStep,
       totalSteps,
-      lossHistory: lossHistoryRaw.map((point) => ({
+      lossHistory: lossHistory.map((point) => ({
         step: point.step,
         loss: point.value,
       })),
-      lrHistory: lrHistoryRaw.map((point) => ({
+      lrHistory: lrHistory.map((point) => ({
         step: point.step,
         lr: point.value,
       })),
-      gradNormHistory: gradNormHistoryRaw.map((point) => ({
+      gradNormHistory: gradNormHistory.map((point) => ({
         step: point.step,
         gradNorm: point.value,
       })),
-      evalLossHistory: evalLossHistoryRaw.map((point) => ({
+      evalLossHistory: evalLossHistory.map((point) => ({
         step: point.step,
         loss: point.value,
       })),
     }),
-    [currentStep, evalLossHistoryRaw, gradNormHistoryRaw, lossHistoryRaw, lrHistoryRaw, totalSteps],
+    [currentStep, evalLossHistory, gradNormHistory, lossHistory, lrHistory, totalSteps],
   );
 
   if (
diff --git a/studio/frontend/src/features/studio/sections/progress-section.tsx b/studio/frontend/src/features/studio/sections/progress-section.tsx
index da89ce66ee..8255283183 100644
--- a/studio/frontend/src/features/studio/sections/progress-section.tsx
+++ b/studio/frontend/src/features/studio/sections/progress-section.tsx
@@ -26,6 +26,7 @@ import {
   useTrainingConfigStore,
   useTrainingRuntimeStore,
 } from "@/features/training";
+import type { TrainingViewData } from "@/features/training";
 import { useGpuUtilization } from "@/hooks";
 import { cn } from "@/lib/utils";
 import {
@@ -39,7 +40,7 @@ import {
 } from "@hugeicons/core-free-icons";
 import { HugeiconsIcon } from "@hugeicons/react";
 import { Link, useNavigate } from "@tanstack/react-router";
-import { type ReactElement, type ReactNode, useEffect, useState } from "react";
+import { type ReactElement, type ReactNode, useState } from "react";
 import { useShallow } from "zustand/react/shallow";
 import { ChartSettingsSheet } from "./charts/chart-settings-sheet";
 import {
@@ -61,34 +62,33 @@ function configRow(
   return [label, value];
 }
 
-export function ProgressSection(): ReactElement {
+interface ProgressSectionProps {
+  data: TrainingViewData;
+  isHistorical?: boolean;
+  configOverride?: {
+    epochs?: number;
+    batchSize?: number;
+    learningRate?: string;
+    maxSteps?: number;
+    contextLength?: number;
+    warmupSteps?: number;
+    optimizerType?: string;
+    loraRank?: number;
+    loraAlpha?: number;
+    loraDropout?: number;
+    loraVariant?: string;
+  };
+}
+
+export function ProgressSection({
+  data,
+  isHistorical = false,
+  configOverride,
+}: ProgressSectionProps): ReactElement {
   const navigate = useNavigate();
-  const runtime = useTrainingRuntimeStore(
-    useShallow((state) => ({
-      phase: state.phase,
-      message: state.message,
-      error: state.error,
-      currentStep: state.currentStep,
-      totalSteps: state.totalSteps,
-      currentEpoch: state.currentEpoch,
-      currentLoss: state.currentLoss,
-      currentLearningRate: state.currentLearningRate,
-      currentGradNorm: state.currentGradNorm,
-      progressPercent: state.progressPercent,
-      elapsedSeconds: state.elapsedSeconds,
-      etaSeconds: state.etaSeconds,
-      currentNumTokens: state.currentNumTokens,
-      isTrainingRunning: state.isTrainingRunning,
-      lossHistory: state.lossHistory,
-      lrHistory: state.lrHistory,
-      gradNormHistory: state.gradNormHistory,
-    })),
-  );
 
   const config = useTrainingConfigStore(
     useShallow((state) => ({
-      selectedModel: state.selectedModel,
-      trainingMethod: state.trainingMethod,
       epochs: state.epochs,
       batchSize: state.batchSize,
       learningRate: state.learningRate,
@@ -103,98 +103,92 @@ export function ProgressSection(): ReactElement {
     })),
   );
 
-  const { stopTrainingRun } = useTrainingActions();
-  const gpu = useGpuUtilization(runtime.isTrainingRunning);
   const [stopDialogOpen, setStopDialogOpen] = useState(false);
-  const [stopRequested, setStopRequested] = useState(false);
+  const [stopRequestedLocal, setStopRequestedLocal] = useState(false);
 
-  useEffect(() => {
-    if (!runtime.isTrainingRunning) {
-      setStopRequested(false);
-    }
-  }, [runtime.isTrainingRunning]);
+  // Auto-reset when training stops -- no useEffect needed
+  const stopRequested = data.isTrainingRunning && stopRequestedLocal;
 
   const pct =
-    runtime.totalSteps > 0
+    data.totalSteps > 0
       ? Math.min(
           100,
           Math.max(
             0,
-            Math.round((runtime.currentStep / runtime.totalSteps) * 100),
+            Math.round((data.currentStep / data.totalSteps) * 100),
           ),
         )
-      : Math.round(runtime.progressPercent);
+      : Math.round(data.progressPercent);
 
-  const elapsed = runtime.elapsedSeconds;
+  const elapsed = data.elapsedSeconds;
   const derivedEta =
     elapsed != null && pct > 0
       ? Math.round((elapsed * (100 - pct)) / Math.max(pct, 1))
       : null;
-  const eta = runtime.etaSeconds ?? derivedEta;
+  const eta = data.etaSeconds ?? derivedEta;
 
   const stepsPerSecond =
-    elapsed != null && elapsed > 0 ? runtime.currentStep / elapsed : null;
+    elapsed != null && elapsed > 0 ? data.currentStep / elapsed : null;
   const showHalfwayHint =
-    runtime.phase === "training" && pct >= 50 && pct < 100;
-  const showCompletedHint = runtime.phase === "completed";
+    data.phase === "training" && pct >= 50 && pct < 100;
+  const showCompletedHint = data.phase === "completed";
   const handleCompareInChat = async () => {
-    setTrainingCompareHandoff(config.selectedModel);
+    setTrainingCompareHandoff(data.modelName);
     await navigate({ to: "/chat" });
   };
-  const requestStop = async (saveCheckpoint: boolean) => {
-    setStopRequested(true);
-    setStopDialogOpen(false);
-    useTrainingRuntimeStore.getState().setStopRequested(true);
-    try {
-      const ok = await stopTrainingRun(saveCheckpoint);
-      if (!ok) {
-        setStopRequested(false);
-      }
-    } catch {
-      setStopRequested(false);
-    }
-  };
 
   const stoppedLoss = getDisplayMetric(
-    runtime.isTrainingRunning,
-    runtime.currentLoss,
-    runtime.lossHistory,
+    data.isTrainingRunning,
+    data.currentLoss,
+    data.lossHistory,
   );
   const stoppedLr = getDisplayMetric(
-    runtime.isTrainingRunning,
-    runtime.currentLearningRate,
-    runtime.lrHistory,
+    data.isTrainingRunning,
+    data.currentLearningRate,
+    data.lrHistory,
   );
-  const stoppedGradNorm = runtime.isTrainingRunning
-    ? runtime.currentGradNorm
-    : (lastNonZeroValue(runtime.gradNormHistory) ?? runtime.currentGradNorm);
+  const stoppedGradNorm = data.isTrainingRunning
+    ? data.currentGradNorm
+    : (lastValue(data.gradNormHistory) ?? data.currentGradNorm);
+
+  const cfgEpochs = isHistorical ? configOverride?.epochs : config.epochs;
+  const cfgBatchSize = isHistorical ? configOverride?.batchSize : config.batchSize;
+  const cfgLearningRate = isHistorical ? configOverride?.learningRate : config.learningRate;
+  const cfgMaxSteps = isHistorical ? configOverride?.maxSteps : config.maxSteps;
+  const cfgContextLength = isHistorical ? configOverride?.contextLength : config.contextLength;
+  const cfgWarmupSteps = isHistorical ? configOverride?.warmupSteps : config.warmupSteps;
+  const cfgOptimizerType = isHistorical ? configOverride?.optimizerType : config.optimizerType;
+  const cfgLoraRank = isHistorical ? configOverride?.loraRank : config.loraRank;
+  const cfgLoraAlpha = isHistorical ? configOverride?.loraAlpha : config.loraAlpha;
+  const cfgLoraDropout = isHistorical ? configOverride?.loraDropout : config.loraDropout;
+  const cfgLoraVariant = isHistorical ? configOverride?.loraVariant : config.loraVariant;
 
   const optimizerLabel =
-    OPTIMIZER_OPTIONS.find((o) => o.value === config.optimizerType)?.label ??
-    config.optimizerType;
+    OPTIMIZER_OPTIONS.find((o) => o.value === cfgOptimizerType)?.label ??
+    cfgOptimizerType;
 
   const configItems: ConfigGroup[] = [
     {
       section: "Hyperparams",
       rows: [
-        configRow("Epochs", config.epochs),
-        configRow("Batch size", config.batchSize),
-        configRow("Learning rate", config.learningRate),
+        configRow("Epochs", cfgEpochs),
+        configRow("Batch size", cfgBatchSize),
+        configRow("Learning rate", cfgLearningRate),
         configRow("Optimizer", optimizerLabel),
-        configRow("Max steps", config.maxSteps),
-        configRow("Context length", config.contextLength),
-        configRow("Warmup steps", config.warmupSteps),
+        configRow("Max steps", cfgMaxSteps),
+        configRow("Context length", cfgContextLength),
+        configRow("Warmup steps", cfgWarmupSteps),
       ],
     },
-    ...(config.trainingMethod !== "full"
+    ...(data.trainingMethod !== "full"
       ? [
           {
             section: "LoRA",
             rows: [
-              configRow("Rank", config.loraRank),
-              configRow("Alpha", config.loraAlpha),
-              configRow("Dropout", config.loraDropout),
-              configRow("Variant", config.loraVariant),
+              configRow("Rank", cfgLoraRank),
+              configRow("Alpha", cfgLoraAlpha),
+              configRow("Dropout", cfgLoraDropout),
+              configRow("Variant", cfgLoraVariant),
             ],
           },
         ]
@@ -205,30 +199,34 @@ export function ProgressSection(): ReactElement {
     <SectionCard
       icon={<HugeiconsIcon icon={ChartAverageIcon} className="size-5" />}
       title="Training Progress"
-      description={runtime.message || "Live training metrics"}
+      description={data.message || "Live training metrics"}
       accent="emerald"
       className="shadow-border border border-border/60 bg-card/90 ring-0 backdrop-blur-sm"
       headerAction={
-        <TrainingHeaderActions
-          configItems={configItems}
-          isTrainingRunning={runtime.isTrainingRunning}
-          onOpenStopDialog={setStopDialogOpen}
-          onRequestStop={requestStop}
-          stopDialogOpen={stopDialogOpen}
-          stopRequested={stopRequested}
-        />
+        isHistorical ? (
+          <ConfigPopoverButton configItems={configItems} />
+        ) : (
+          <LiveTrainingHeaderActions
+            configItems={configItems}
+            isTrainingRunning={data.isTrainingRunning}
+            onOpenStopDialog={setStopDialogOpen}
+            stopDialogOpen={stopDialogOpen}
+            stopRequested={stopRequested}
+            onSetStopRequested={setStopRequestedLocal}
+          />
+        )
       }
     >
       <div className="grid grid-cols-1 gap-5 lg:grid-cols-[minmax(0,1.2fr)_minmax(18rem,0.8fr)]">
         <div className="flex flex-col gap-4">
           <div className="flex flex-wrap items-center gap-2">
             <span
-              className={`rounded-full px-2.5 py-1 text-[10px] font-semibold ${phaseColors[runtime.phase]}`}
+              className={`rounded-full px-2.5 py-1 text-[10px] font-semibold ${phaseColors[data.phase]}`}
             >
-              {phaseLabel[runtime.phase]}
+              {phaseLabel[data.phase]}
             </span>
             <span className="text-[10px] tabular-nums text-muted-foreground">
-              Epoch {runtime.currentEpoch.toFixed(2)}
+              Epoch {formatNumber(data.currentEpoch, 2)}
             </span>
             <span className="rounded-full border border-border/60 px-2.5 py-1 text-[10px] font-medium tabular-nums text-muted-foreground">
               {pct}% complete
@@ -238,22 +236,24 @@ export function ProgressSection(): ReactElement {
           <div className="flex flex-col gap-2">
             <div className="flex justify-between text-xs text-muted-foreground">
               <span>
-                Step {runtime.currentStep} / {runtime.totalSteps || "--"}
+                Step {data.currentStep} / {data.totalSteps || "--"}
               </span>
               <span>{pct}%</span>
             </div>
             <Progress value={pct} className="h-2 bg-foreground/[0.05]" />
           </div>
 
-          <MilestoneCallout
-            showCompletedHint={showCompletedHint}
-            showHalfwayHint={showHalfwayHint}
-            onCompareInChat={handleCompareInChat}
-          />
+          {!isHistorical && (
+            <MilestoneCallout
+              showCompletedHint={showCompletedHint}
+              showHalfwayHint={showHalfwayHint}
+              onCompareInChat={handleCompareInChat}
+            />
+          )}
 
-          {runtime.error && (
+          {data.error && (
             <p className="rounded-2xl border border-destructive/30 bg-destructive/5 px-3 py-2 text-xs text-red-500 leading-relaxed">
-              {runtime.error}
+              {data.error}
             </p>
           )}
 
@@ -262,97 +262,196 @@ export function ProgressSection(): ReactElement {
               label="Loss"
               valueClassName="text-2xl font-bold tracking-tight"
             >
-              {stoppedLoss.toFixed(4)}
+              {stoppedLoss != null ? stoppedLoss.toFixed(4) : "--"}
             </MetricStat>
-            <MetricStat label="LR">{stoppedLr.toExponential(2)}</MetricStat>
+            <MetricStat label="LR">{stoppedLr != null ? stoppedLr.toExponential(2) : "--"}</MetricStat>
             <MetricStat label="Grad Norm">
               {formatNumber(stoppedGradNorm, 3)}
             </MetricStat>
             <MetricStat label="Model" valueClassName="truncate">
-              {config.selectedModel ?? "--"}
+              {data.modelName || "--"}
             </MetricStat>
             <MetricStat label="Method">
-              {config.trainingMethod === "qlora" ? "QLoRA" : config.trainingMethod === "lora" ? "LoRA" : "Full"}
+              {data.trainingMethod === "qlora" ? "QLoRA" : data.trainingMethod === "lora" ? "LoRA" : "Full"}
             </MetricStat>
           </div>
 
           <div className="flex flex-wrap gap-x-4 gap-y-1 text-xs text-muted-foreground">
             <span>Elapsed: {formatDuration(elapsed)}</span>
-            <span>ETA: {formatDuration(eta)}</span>
+            {!isHistorical && <span>ETA: {formatDuration(eta)}</span>}
             <span>
               {stepsPerSecond == null
                 ? "-- steps/s"
                 : `${stepsPerSecond.toFixed(2)} steps/s`}
             </span>
-            {runtime.currentNumTokens != null && (
-              <span>Tokens: {runtime.currentNumTokens}</span>
+            {data.currentNumTokens != null && (
+              <span>Tokens: {data.currentNumTokens}</span>
             )}
           </div>
         </div>
 
-        <div className="flex flex-col gap-3">
-          <div className="flex items-center justify-between">
-            <p className="text-xs font-medium text-muted-foreground">
-              GPU Monitor
-            </p>
-            <span className="text-[11px] text-muted-foreground">Live</span>
-          </div>
-          <div className="grid grid-cols-2 gap-2.5">
-            <GpuStat
-              label="Utilization"
-              icon={
-                <HugeiconsIcon
-                  icon={DashboardSpeed01Icon}
-                  className="size-3.5"
-                />
-              }
-              value={
-                gpu.gpu_utilization_pct != null
-                  ? `${gpu.gpu_utilization_pct}%`
-                  : "--"
-              }
-              pct={gpu.gpu_utilization_pct ?? 0}
-            />
-            <GpuStat
-              label="Temperature"
-              icon={
-                <HugeiconsIcon icon={TemperatureIcon} className="size-3.5" />
-              }
-              value={
-                gpu.temperature_c != null ? `${gpu.temperature_c}°C` : "--"
-              }
-              pct={gpu.temperature_c ?? 0}
-              max={100}
-            />
-            <GpuStat
-              label="VRAM"
-              icon={<HugeiconsIcon icon={RamMemoryIcon} className="size-3.5" />}
-              value={
-                gpu.vram_used_gb != null && gpu.vram_total_gb != null
-                  ? `${gpu.vram_used_gb} / ${gpu.vram_total_gb} GB`
-                  : "--"
-              }
-              pct={gpu.vram_utilization_pct ?? 0}
-            />
-            <GpuStat
-              label="Power"
-              icon={<HugeiconsIcon icon={ZapIcon} className="size-3.5" />}
-              value={
-                gpu.power_draw_w != null
-                  ? gpu.power_limit_w != null
-                    ? `${gpu.power_draw_w} / ${gpu.power_limit_w} W`
-                    : `${gpu.power_draw_w} W`
-                  : "--"
-              }
-              pct={gpu.power_utilization_pct ?? 0}
-            />
-          </div>
-        </div>
+        {!isHistorical && (
+          <LiveGpuPanel isTrainingRunning={data.isTrainingRunning} />
+        )}
       </div>
     </SectionCard>
   );
 }
 
+function LiveGpuPanel({
+  isTrainingRunning,
+}: {
+  isTrainingRunning: boolean;
+}): ReactElement {
+  const gpu = useGpuUtilization(isTrainingRunning);
+
+  return (
+    <div className="flex flex-col gap-3">
+      <div className="flex items-center justify-between">
+        <p className="text-xs font-medium text-muted-foreground">
+          GPU Monitor
+        </p>
+        <span className="text-[11px] text-muted-foreground">Live</span>
+      </div>
+      <div className="grid grid-cols-2 gap-2.5">
+        <GpuStat
+          label="Utilization"
+          icon={
+            <HugeiconsIcon
+              icon={DashboardSpeed01Icon}
+              className="size-3.5"
+            />
+          }
+          value={
+            gpu.gpu_utilization_pct != null
+              ? `${gpu.gpu_utilization_pct}%`
+              : "--"
+          }
+          pct={gpu.gpu_utilization_pct ?? 0}
+        />
+        <GpuStat
+          label="Temperature"
+          icon={
+            <HugeiconsIcon icon={TemperatureIcon} className="size-3.5" />
+          }
+          value={
+            gpu.temperature_c != null ? `${gpu.temperature_c}°C` : "--"
+          }
+          pct={gpu.temperature_c ?? 0}
+          max={100}
+        />
+        <GpuStat
+          label="VRAM"
+          icon={<HugeiconsIcon icon={RamMemoryIcon} className="size-3.5" />}
+          value={
+            gpu.vram_used_gb != null && gpu.vram_total_gb != null
+              ? `${gpu.vram_used_gb} / ${gpu.vram_total_gb} GB`
+              : "--"
+          }
+          pct={gpu.vram_utilization_pct ?? 0}
+        />
+        <GpuStat
+          label="Power"
+          icon={<HugeiconsIcon icon={ZapIcon} className="size-3.5" />}
+          value={
+            gpu.power_draw_w != null
+              ? gpu.power_limit_w != null
+                ? `${gpu.power_draw_w} / ${gpu.power_limit_w} W`
+                : `${gpu.power_draw_w} W`
+              : "--"
+          }
+          pct={gpu.power_utilization_pct ?? 0}
+        />
+      </div>
+    </div>
+  );
+}
+
+function LiveTrainingHeaderActions({
+  configItems,
+  isTrainingRunning,
+  onOpenStopDialog,
+  stopDialogOpen,
+  stopRequested,
+  onSetStopRequested,
+}: {
+  configItems: ConfigGroup[];
+  isTrainingRunning: boolean;
+  onOpenStopDialog: (open: boolean) => void;
+  stopDialogOpen: boolean;
+  stopRequested: boolean;
+  onSetStopRequested: (v: boolean) => void;
+}): ReactElement {
+  const { stopTrainingRun } = useTrainingActions();
+
+  const requestStop = async (saveCheckpoint: boolean) => {
+    onSetStopRequested(true);
+    onOpenStopDialog(false);
+    useTrainingRuntimeStore.getState().setStopRequested(true);
+    try {
+      const ok = await stopTrainingRun(saveCheckpoint);
+      if (!ok) {
+        onSetStopRequested(false);
+      }
+    } catch {
+      onSetStopRequested(false);
+    }
+  };
+
+  return (
+    <TrainingHeaderActions
+      configItems={configItems}
+      isTrainingRunning={isTrainingRunning}
+      onOpenStopDialog={onOpenStopDialog}
+      onRequestStop={requestStop}
+      stopDialogOpen={stopDialogOpen}
+      stopRequested={stopRequested}
+    />
+  );
+}
+
+function ConfigPopoverButton({
+  configItems,
+}: {
+  configItems: ConfigGroup[];
+}): ReactElement {
+  return (
+    <Popover>
+      <PopoverTrigger asChild={true}>
+        <Button
+          type="button"
+          variant="ghost"
+          size="icon-sm"
+          className="rounded-full text-muted-foreground hover:bg-muted hover:text-foreground"
+          aria-label="Open training config"
+        >
+          <HugeiconsIcon icon={Notebook01Icon} className="size-4" />
+        </Button>
+      </PopoverTrigger>
+      <PopoverContent className="w-72" align="end">
+        <div className="flex flex-col gap-3">
+          <p className="text-xs font-semibold">Training Config</p>
+          {configItems.map((group) => (
+            <div key={group.section} className="flex flex-col gap-1">
+              <p className="text-[10px] font-semibold uppercase tracking-wider text-muted-foreground">
+                {group.section}
+              </p>
+              {group.rows.map(([label, value]) => (
+                <div key={label} className="flex justify-between text-xs">
+                  <span className="text-muted-foreground">{label}</span>
+                  <span className="font-medium tabular-nums">
+                    {value == null || value === "" ? "--" : String(value)}
+                  </span>
+                </div>
+              ))}
+            </div>
+          ))}
+        </div>
+      </PopoverContent>
+    </Popover>
+  );
+}
+
 function TrainingHeaderActions({
   configItems,
   isTrainingRunning,
@@ -370,39 +469,7 @@ function TrainingHeaderActions({
 }): ReactElement {
   return (
     <div className="flex items-center gap-2">
-      <Popover>
-        <PopoverTrigger asChild={true}>
-          <Button
-            type="button"
-            variant="ghost"
-            size="icon-sm"
-            className="rounded-full text-muted-foreground hover:bg-muted hover:text-foreground"
-            aria-label="Open training config"
-          >
-            <HugeiconsIcon icon={Notebook01Icon} className="size-4" />
-          </Button>
-        </PopoverTrigger>
-        <PopoverContent className="w-72" align="end">
-          <div className="flex flex-col gap-3">
-            <p className="text-xs font-semibold">Training Config</p>
-            {configItems.map((group) => (
-              <div key={group.section} className="flex flex-col gap-1">
-                <p className="text-[10px] font-semibold uppercase tracking-wider text-muted-foreground">
-                  {group.section}
-                </p>
-                {group.rows.map(([label, value]) => (
-                  <div key={label} className="flex justify-between text-xs">
-                    <span className="text-muted-foreground">{label}</span>
-                    <span className="font-medium tabular-nums">
-                      {String(value)}
-                    </span>
-                  </div>
-                ))}
-              </div>
-            ))}
-          </div>
-        </PopoverContent>
-      </Popover>
+      <ConfigPopoverButton configItems={configItems} />
       <ChartSettingsSheet />
       <AlertDialog open={stopDialogOpen} onOpenChange={onOpenStopDialog}>
         <Button
@@ -518,25 +585,21 @@ function MetricStat({
   );
 }
 
-function lastNonZeroValue(points: { value: number }[]): number | null {
-  for (let i = points.length - 1; i >= 0; i -= 1) {
-    const value = points[i]?.value;
-    if (Number.isFinite(value) && value !== 0) {
-      return value;
-    }
-  }
-  return null;
+function lastValue(points: { value: number }[]): number | null {
+  if (points.length === 0) return null;
+  const v = points[points.length - 1]?.value;
+  return v != null && Number.isFinite(v) ? v : null;
 }
 
 function getDisplayMetric(
   isTrainingRunning: boolean,
-  currentValue: number,
+  currentValue: number | null,
   history: { value: number }[],
-): number {
+): number | null {
   if (isTrainingRunning) {
-    return currentValue;
+    return currentValue != null ? currentValue : null;
   }
-  return lastNonZeroValue(history) ?? currentValue;
+  return lastValue(history) ?? (currentValue != null ? currentValue : null);
 }
 
 function GpuStat({
diff --git a/studio/frontend/src/features/studio/studio-page.tsx b/studio/frontend/src/features/studio/studio-page.tsx
index 98aba62e73..a46e09cf7b 100644
--- a/studio/frontend/src/features/studio/studio-page.tsx
+++ b/studio/frontend/src/features/studio/studio-page.tsx
@@ -1,26 +1,28 @@
 // SPDX-License-Identifier: AGPL-3.0-only
 // Copyright 2026-present the Unsloth AI Inc. team. All rights reserved. See /studio/LICENSE.AGPL-3.0
 
-import { Button } from "@/components/ui/button";
 import {
   shouldShowTrainingView,
   useDatasetPreviewDialogStore,
-  useTrainingActions,
   useTrainingConfigStore,
   useTrainingRuntimeLifecycle,
   useTrainingRuntimeStore,
 } from "@/features/training";
 import { GuidedTour, useGuidedTourController } from "@/features/tour";
 import { studioTourSteps, studioTrainingTourSteps } from "./tour";
+import { Tabs, TabsContent, TabsList, TabsTrigger } from "@/components/ui/tabs";
+import { Button } from "@/components/ui/button";
 import { ArrowLeft01Icon } from "@hugeicons/core-free-icons";
 import { HugeiconsIcon } from "@hugeicons/react";
-import { type ReactElement, useEffect } from "react";
+import { type ReactElement, useEffect, useState } from "react";
 import { DatasetPreviewDialog } from "./sections/dataset-preview-dialog";
 import { DatasetSection } from "./sections/dataset-section";
 import { ModelSection } from "./sections/model-section";
 import { ParamsSection } from "./sections/params-section";
 import { TrainingSection } from "./sections/training-section";
-import { TrainingView } from "./training-view";
+import { LiveTrainingView } from "./live-training-view";
+import { HistoricalTrainingView } from "./historical-training-view";
+import { HistoryCardGrid } from "./history-card-grid";
 
 const STUDIO_TOUR_KEY = "tour:studio:v1";
 
@@ -28,11 +30,10 @@ export function StudioPage(): ReactElement {
   useTrainingRuntimeLifecycle();
   const showTrainingView = useTrainingRuntimeStore(shouldShowTrainingView);
   const isTrainingRunning = useTrainingRuntimeStore((state) => state.isTrainingRunning);
+  const currentJobId = useTrainingRuntimeStore((state) => state.jobId);
   const runtimeMessage = useTrainingRuntimeStore((state) => state.message);
-  const runtimePhase = useTrainingRuntimeStore((state) => state.phase);
   const isHydratingRuntime = useTrainingRuntimeStore((state) => state.isHydrating);
   const hasHydratedRuntime = useTrainingRuntimeStore((state) => state.hasHydrated);
-  const { dismissTrainingRun } = useTrainingActions();
 
   const config = useTrainingConfigStore();
   const selectedModel = useTrainingConfigStore((s) => s.selectedModel);
@@ -47,19 +48,22 @@ export function StudioPage(): ReactElement {
   const dialogInitial = useDatasetPreviewDialogStore((s) => s.initialData);
   const closeDialog = useDatasetPreviewDialogStore((s) => s.close);
 
-  const stopRequested = useTrainingRuntimeStore((state) => state.stopRequested);
-  const canGoBack =
-    showTrainingView &&
-    !isHydratingRuntime &&
-    (stopRequested ||
-      (!isTrainingRunning &&
-        (runtimePhase === "stopped" ||
-          runtimePhase === "error" ||
-          runtimePhase === "completed" ||
-          runtimePhase === "idle")));
+  const [requestedTab, setRequestedTab] = useState("configure");
+  const [selectedHistoryRunId, setSelectedHistoryRunId] = useState<string | null>(null);
+
+  // Derive activeTab: auto-switch to "current-run" only while training is
+  // genuinely running.  Once training ends, honour whatever tab the user clicks.
+  // If requestedTab is "current-run" but there's nothing to show, fall back to "configure".
+  const activeTab =
+    isTrainingRunning && requestedTab !== "history"
+      ? "current-run"
+      : requestedTab === "current-run" && !showTrainingView
+        ? "configure"
+        : requestedTab;
+
   const tourEnabled = hasHydratedRuntime && !isHydratingRuntime;
-  const isConfigTour = !showTrainingView;
-  const tourSteps = showTrainingView ? studioTrainingTourSteps : studioTourSteps;
+  const isConfigTour = activeTab === "configure";
+  const tourSteps = activeTab === "current-run" ? studioTrainingTourSteps : studioTourSteps;
   const tour = useGuidedTourController({
     id: "studio",
     steps: tourSteps,
@@ -71,13 +75,36 @@ export function StudioPage(): ReactElement {
   const setTourOpen = tour.setOpen;
   useEffect(() => {
     setTourOpen(false);
-  }, [showTrainingView, setTourOpen]);
+  }, [activeTab, setTourOpen]);
+
+  // When training auto-switches us to "current-run", persist that in
+  // requestedTab so the user stays on results after training ends.
+  useEffect(() => {
+    if (isTrainingRunning && requestedTab !== "history" && requestedTab !== "current-run") {
+      setRequestedTab("current-run");
+      setSelectedHistoryRunId(null);
+    }
+  }, [isTrainingRunning, requestedTab]);
 
   useEffect(() => {
     ensureModelDefaultsLoaded();
     ensureDatasetChecked();
   }, [selectedModel, ensureModelDefaultsLoaded, ensureDatasetChecked]);
 
+  function handleTabChange(value: string) {
+    setRequestedTab(value);
+    if (value !== "history") {
+      setSelectedHistoryRunId(null);
+    }
+  }
+
+  const subtitle = (() => {
+    if (activeTab === "current-run") return runtimeMessage || "Training in progress";
+    if (activeTab === "history")
+      return selectedHistoryRunId ? "Viewing past run" : "View past training runs";
+    return "Configure and start training";
+  })();
+
   return (
     <div className="relative min-h-screen overflow-hidden bg-background">
       <main className="relative z-10 mx-auto max-w-7xl px-4 py-4 sm:px-6">
@@ -100,42 +127,69 @@ export function StudioPage(): ReactElement {
           isVlm={config.isVisionModel && config.isDatasetImage === true}
         />
 
-        {canGoBack && (
-          <Button
-            variant="ghost"
-            size="sm"
-            className="mb-2 cursor-pointer gap-1.5 text-muted-foreground"
-            onClick={() => void dismissTrainingRun()}
-          >
-            <HugeiconsIcon icon={ArrowLeft01Icon} className="size-4" />
-            Back to configuration
-          </Button>
-        )}
-
         <div className="mb-6 flex flex-col gap-0.5 sm:mb-8">
           <h1 className="text-2xl font-semibold tracking-tight">
             Fine-tuning Studio
           </h1>
-          <p className="text-sm text-muted-foreground">
-            {showTrainingView
-              ? runtimeMessage || "Training in progress"
-              : "Configure and start training"}
-          </p>
+          <p className="text-sm text-muted-foreground">{subtitle}</p>
         </div>
 
         {!hasHydratedRuntime && isHydratingRuntime ? (
           <div className="rounded-xl border bg-card p-8 text-sm text-muted-foreground">
             Loading training runtime...
           </div>
-        ) : showTrainingView ? (
-          <TrainingView />
         ) : (
-          <div className="grid grid-cols-1 items-start gap-4 md:grid-cols-2 md:gap-6 xl:grid-cols-12">
-            <ModelSection />
-            <DatasetSection />
-            <ParamsSection />
-            <TrainingSection />
-          </div>
+          <Tabs value={activeTab} onValueChange={handleTabChange}>
+            <div className="flex items-center gap-3">
+              {selectedHistoryRunId && activeTab === "history" && (
+                <Button
+                  variant="ghost"
+                  size="icon-sm"
+                  className="rounded-full text-muted-foreground"
+                  onClick={() => setSelectedHistoryRunId(null)}
+                  aria-label="Back to history"
+                >
+                  <HugeiconsIcon icon={ArrowLeft01Icon} className="size-4" />
+                </Button>
+              )}
+              <TabsList variant="line">
+                <TabsTrigger value="configure" disabled={isTrainingRunning}>
+                  Configure
+                </TabsTrigger>
+                <TabsTrigger value="current-run" disabled={!showTrainingView}>
+                  Current Run
+                </TabsTrigger>
+                <TabsTrigger value="history">History</TabsTrigger>
+              </TabsList>
+            </div>
+
+            <TabsContent value="configure">
+              <div className="grid grid-cols-1 items-start gap-4 md:grid-cols-2 md:gap-6 xl:grid-cols-12">
+                <ModelSection />
+                <DatasetSection />
+                <ParamsSection />
+                <TrainingSection />
+              </div>
+            </TabsContent>
+
+            <TabsContent value="current-run">
+              <LiveTrainingView />
+            </TabsContent>
+
+            <TabsContent value="history">
+              {selectedHistoryRunId ? (
+                <HistoricalTrainingView runId={selectedHistoryRunId} />
+              ) : (
+                <HistoryCardGrid onSelectRun={(runId) => {
+                  if (runId === currentJobId && isTrainingRunning) {
+                    handleTabChange("current-run");
+                  } else {
+                    setSelectedHistoryRunId(runId);
+                  }
+                }} />
+              )}
+            </TabsContent>
+          </Tabs>
         )}
       </main>
     </div>
diff --git a/studio/frontend/src/features/studio/training-view.tsx b/studio/frontend/src/features/studio/training-view.tsx
deleted file mode 100644
index b995b2f528..0000000000
--- a/studio/frontend/src/features/studio/training-view.tsx
+++ /dev/null
@@ -1,57 +0,0 @@
-// SPDX-License-Identifier: AGPL-3.0-only
-// Copyright 2026-present the Unsloth AI Inc. team. All rights reserved. See /studio/LICENSE.AGPL-3.0
-
-import { cn } from "@/lib/utils";
-import { useTrainingRuntimeStore } from "@/features/training";
-import type { ReactElement } from "react";
-import { useShallow } from "zustand/react/shallow";
-import { ChartsSection } from "./sections/charts-section";
-import { ProgressSection } from "./sections/progress-section";
-import { TrainingStartOverlay } from "./training-start-overlay";
-
-export function TrainingView(): ReactElement {
-  const runtime = useTrainingRuntimeStore(
-    useShallow((state) => ({
-      phase: state.phase,
-      message: state.message,
-      currentStep: state.currentStep,
-      firstStepReceived: state.firstStepReceived,
-      isStarting: state.isStarting,
-    })),
-  );
-
-  const isPreparingPhase =
-    runtime.phase === "downloading_model" ||
-    runtime.phase === "downloading_dataset" ||
-    runtime.phase === "loading_model" ||
-    runtime.phase === "loading_dataset" ||
-    runtime.phase === "configuring";
-  const isWaitingForFirstStep =
-    runtime.phase === "training" && !runtime.firstStepReceived;
-  const showOverlay =
-    runtime.isStarting ||
-    isPreparingPhase ||
-    (isWaitingForFirstStep && runtime.currentStep <= 0);
-
-  return (
-    <div className={cn("relative", showOverlay && "min-h-[72vh]")}>
-      <div
-        className={cn(
-          "relative z-10 flex flex-col gap-6 transition-[filter]",
-          showOverlay && "blur",
-        )}
-      >
-        <div data-tour="studio-training-progress">
-          <ProgressSection />
-        </div>
-        <ChartsSection />
-      </div>
-      {showOverlay ? (
-        <TrainingStartOverlay
-          message={runtime.message}
-          currentStep={runtime.currentStep}
-        />
-      ) : null}
-    </div>
-  );
-}
diff --git a/studio/frontend/src/features/training/api/history-api.ts b/studio/frontend/src/features/training/api/history-api.ts
new file mode 100644
index 0000000000..8f279eb439
--- /dev/null
+++ b/studio/frontend/src/features/training/api/history-api.ts
@@ -0,0 +1,59 @@
+// SPDX-License-Identifier: AGPL-3.0-only
+// Copyright 2026-present the Unsloth AI Inc. team. All rights reserved. See /studio/LICENSE.AGPL-3.0
+
+import { authFetch } from "@/features/auth";
+import type {
+  TrainingRunDeleteResponse,
+  TrainingRunDetailResponse,
+  TrainingRunListResponse,
+} from "../types/history";
+
+async function readError(response: Response): Promise<string> {
+  try {
+    const payload = (await response.json()) as { detail?: string; message?: string };
+    return payload.detail || payload.message || `Request failed (${response.status})`;
+  } catch {
+    return `Request failed (${response.status})`;
+  }
+}
+
+async function parseJson<T>(response: Response): Promise<T> {
+  if (!response.ok) {
+    throw new Error(await readError(response));
+  }
+  return (await response.json()) as T;
+}
+
+export async function listTrainingRuns(
+  limit = 50,
+  offset = 0,
+  signal?: AbortSignal,
+): Promise<TrainingRunListResponse> {
+  const response = await authFetch(
+    `/api/train/runs?limit=${limit}&offset=${offset}`,
+    { signal },
+  );
+  return parseJson<TrainingRunListResponse>(response);
+}
+
+export async function getTrainingRun(
+  runId: string,
+  signal?: AbortSignal,
+): Promise<TrainingRunDetailResponse> {
+  const response = await authFetch(
+    `/api/train/runs/${encodeURIComponent(runId)}`,
+    { signal },
+  );
+  return parseJson<TrainingRunDetailResponse>(response);
+}
+
+export async function deleteTrainingRun(
+  runId: string,
+  signal?: AbortSignal,
+): Promise<TrainingRunDeleteResponse> {
+  const response = await authFetch(
+    `/api/train/runs/${encodeURIComponent(runId)}`,
+    { method: "DELETE", signal },
+  );
+  return parseJson<TrainingRunDeleteResponse>(response);
+}
diff --git a/studio/frontend/src/features/training/index.ts b/studio/frontend/src/features/training/index.ts
index 9a70e0a71c..af34b306cf 100644
--- a/studio/frontend/src/features/training/index.ts
+++ b/studio/frontend/src/features/training/index.ts
@@ -14,6 +14,14 @@ export { useDatasetPreviewDialogStore } from "./stores/dataset-preview-dialog-st
 export { uploadTrainingDataset } from "./api/datasets-api";
 export { listLocalModels } from "./api/models-api";
 export type { LocalModelInfo } from "./api/models-api";
-export type { TrainingPhase } from "./types/runtime";
+export type { TrainingPhase, TrainingViewData, TrainingSeriesPoint } from "./types/runtime";
+export type {
+  TrainingRunSummary,
+  TrainingRunListResponse,
+  TrainingRunMetrics,
+  TrainingRunDetailResponse,
+  TrainingRunDeleteResponse,
+} from "./types/history";
+export { listTrainingRuns, getTrainingRun, deleteTrainingRun } from "./api/history-api";
 export { parseYamlConfig, serializeConfigToYaml } from "./lib/yaml-config";
 export { validateTrainingConfig } from "./lib/validation";
diff --git a/studio/frontend/src/features/training/types/history.ts b/studio/frontend/src/features/training/types/history.ts
new file mode 100644
index 0000000000..8b89db539b
--- /dev/null
+++ b/studio/frontend/src/features/training/types/history.ts
@@ -0,0 +1,48 @@
+// SPDX-License-Identifier: AGPL-3.0-only
+// Copyright 2026-present the Unsloth AI Inc. team. All rights reserved. See /studio/LICENSE.AGPL-3.0
+
+export interface TrainingRunSummary {
+  id: string;
+  status: "running" | "completed" | "stopped" | "error";
+  model_name: string;
+  dataset_name: string;
+  started_at: string;
+  ended_at: string | null;
+  total_steps: number | null;
+  final_step: number | null;
+  final_loss: number | null;
+  output_dir: string | null;
+  duration_seconds: number | null;
+  error_message: string | null;
+  loss_sparkline: number[] | null;
+}
+
+export interface TrainingRunListResponse {
+  runs: TrainingRunSummary[];
+  total: number;
+}
+
+export interface TrainingRunMetrics {
+  step_history: number[];
+  loss_history: number[];
+  loss_step_history: number[];
+  lr_history: number[];
+  lr_step_history: number[];
+  grad_norm_history: number[];
+  grad_norm_step_history: number[];
+  eval_loss_history: number[];
+  eval_step_history: number[];
+  final_epoch: number | null;
+  final_num_tokens: number | null;
+}
+
+export interface TrainingRunDetailResponse {
+  run: TrainingRunSummary;
+  config: Record<string, unknown>;
+  metrics: TrainingRunMetrics;
+}
+
+export interface TrainingRunDeleteResponse {
+  status: string;
+  message: string;
+}
diff --git a/studio/frontend/src/features/training/types/runtime.ts b/studio/frontend/src/features/training/types/runtime.ts
index 7669c8b2f3..1bf319a5d1 100644
--- a/studio/frontend/src/features/training/types/runtime.ts
+++ b/studio/frontend/src/features/training/types/runtime.ts
@@ -53,8 +53,8 @@ export interface TrainingProgressPayload {
   job_id: string;
   step: number;
   total_steps: number;
-  loss: number;
-  learning_rate: number;
+  loss: number | null;
+  learning_rate: number | null;
   progress_percent: number;
   epoch: number | null;
   elapsed_seconds: number | null;
@@ -118,3 +118,32 @@ export interface TrainingRuntimeActions {
 }
 
 export type TrainingRuntimeStore = TrainingRuntimeState & TrainingRuntimeActions;
+
+export interface TrainingViewData {
+  // Current metrics (for ProgressSection)
+  phase: TrainingPhase;
+  currentStep: number;
+  totalSteps: number;
+  currentLoss: number | null;
+  currentLearningRate: number | null;
+  currentGradNorm: number | null;
+  currentEpoch: number | null;
+  currentNumTokens: number | null;
+  progressPercent: number;
+  elapsedSeconds: number | null;
+  etaSeconds: number | null;
+  evalEnabled: boolean;
+  message: string;
+  error: string | null;
+  isTrainingRunning: boolean;
+
+  // Config summary
+  modelName: string;
+  trainingMethod: string;
+
+  // Time-series (for ChartsSection)
+  lossHistory: TrainingSeriesPoint[];
+  lrHistory: TrainingSeriesPoint[];
+  gradNormHistory: TrainingSeriesPoint[];
+  evalLossHistory: TrainingSeriesPoint[];
+}