feat(core): add Snowflake Cortex provider (#29901)

Co-authored-by: Cortex Code <noreply@snowflake.com>

packages/core/src/plugin/provider.ts

@@ -19,6 +19,7 @@ import { LLMGatewayPlugin } from "./provider/llmgateway"
 import { MistralPlugin } from "./provider/mistral"
 import { NvidiaPlugin } from "./provider/nvidia"
 import { OpenAIPlugin } from "./provider/openai"
+import { SnowflakeCortexPlugin } from "./provider/snowflake-cortex"
 import { OpenAICompatiblePlugin } from "./provider/openai-compatible"
 import { OpencodePlugin } from "./provider/opencode"
 import { OpenRouterPlugin } from "./provider/openrouter"
@@ -53,6 +54,7 @@ export const ProviderPlugins = [
   MistralPlugin,
   NvidiaPlugin,
   OpencodePlugin,
+  SnowflakeCortexPlugin,
   OpenAICompatiblePlugin,
   OpenAIPlugin,
   OpenRouterPlugin,

packages/core/src/plugin/provider/snowflake-cortex.ts

@@ -0,0 +1,81 @@
+import { Effect } from "effect"
+import { PluginV2 } from "../../plugin"
+import { ProviderV2 } from "../../provider"
+
+type FetchLike = (url: string | URL | Request, init?: RequestInit) => Promise<Response>
+
+// Exported for testing: intercepts Cortex-specific request/response quirks.
+export function cortexFetch(upstream: FetchLike = fetch) {
+  return async (url: string | URL | Request, init?: RequestInit): Promise<Response> => {
+    if (init?.body && typeof init.body === "string") {
+      try {
+        const body = JSON.parse(init.body)
+        if ("max_tokens" in body) {
+          body.max_completion_tokens = body.max_tokens
+          delete body.max_tokens
+          init = { ...init, body: JSON.stringify(body) }
+        }
+      } catch {}
+    }
+
+    const response = await upstream(url, init)
+
+    // Cortex returns 400 "conversation complete" as a normal stop condition
+    if (!response.ok && response.status === 400) {
+      try {
+        const errorData = (await response.clone().json()) as Record<string, unknown>
+        if (String(errorData.message || errorData.error || "").toLowerCase().includes("conversation complete")) {
+          return new Response(
+            JSON.stringify({ choices: [{ finish_reason: "stop", message: { content: "", role: "assistant" } }] }),
+            { status: 200, headers: new Headers({ "content-type": "application/json" }) },
+          )
+        }
+      } catch {}
+    }
+
+    // Cortex returns role:"" in streaming deltas; the AI SDK schema requires "assistant"
+    if (response.body && response.headers.get("content-type")?.includes("text/event-stream")) {
+      const reader = response.body.getReader()
+      const encoder = new TextEncoder()
+      const decoder = new TextDecoder()
+      const stream = new ReadableStream({
+        async pull(ctrl) {
+          const { done, value } = await reader.read()
+          if (done) {
+            ctrl.close()
+            return
+          }
+          ctrl.enqueue(encoder.encode(decoder.decode(value, { stream: true }).replace(/"role"\s*:\s*""/g, '"role":"assistant"')))
+        },
+        cancel() {
+          reader.cancel()
+        },
+      })
+      return new Response(stream, { headers: response.headers, status: response.status })
+    }
+
+    return response
+  }
+}
+
+export const SnowflakeCortexPlugin = PluginV2.define({
+  id: PluginV2.ID.make("snowflake-cortex"),
+  effect: Effect.gen(function* () {
+    return {
+      "aisdk.sdk": Effect.fn(function* (evt) {
+        if (evt.model.providerID !== ProviderV2.ID.make("snowflake-cortex")) return
+        const pat =
+          process.env.SNOWFLAKE_CORTEX_PAT ??
+          (typeof evt.options.apiKey === "string" ? evt.options.apiKey : undefined)
+        const upstream = typeof evt.options.fetch === "function" ? (evt.options.fetch as FetchLike) : undefined
+        if (evt.options.includeUsage !== false) evt.options.includeUsage = true
+        const mod = yield* Effect.promise(() => import("@ai-sdk/openai-compatible"))
+        evt.sdk = mod.createOpenAICompatible({
+          ...evt.options,
+          ...(pat ? { apiKey: pat } : {}),
+          fetch: cortexFetch(upstream) as typeof fetch,
+        } as any)
+      }),
+    }
+  }),
+})

packages/core/test/plugin/provider-snowflake-cortex.test.ts

@@ -0,0 +1,185 @@
+import { describe, expect, it as bun_it } from "bun:test"
+import { Effect } from "effect"
+import { PluginV2 } from "@opencode-ai/core/plugin"
+import { SnowflakeCortexPlugin, cortexFetch } from "@opencode-ai/core/plugin/provider/snowflake-cortex"
+import { ProviderPlugins } from "@opencode-ai/core/plugin/provider"
+import { expectPluginRegistered, it, model, withEnv } from "./provider-helper"
+
+describe("SnowflakeCortexPlugin", () => {
+  it.effect("is registered in ProviderPlugins before OpenAICompatiblePlugin", () =>
+    Effect.sync(() => {
+      expectPluginRegistered(
+        ProviderPlugins.map((item) => item.id),
+        "snowflake-cortex",
+      )
+      const ids = ProviderPlugins.map((p) => p.id as string)
+      expect(ids.indexOf("snowflake-cortex")).toBeLessThan(ids.indexOf("openai-compatible"))
+    }),
+  )
+
+  it.effect("ignores non-snowflake-cortex providers", () =>
+    Effect.gen(function* () {
+      const plugin = yield* PluginV2.Service
+      yield* plugin.add(SnowflakeCortexPlugin)
+      const result = yield* plugin.trigger(
+        "aisdk.sdk",
+        { model: model("openai", "gpt-4"), package: "@ai-sdk/openai", options: { name: "openai" } },
+        {},
+      )
+      expect(result.sdk).toBeUndefined()
+    }),
+  )
+
+  it.effect("creates SDK for snowflake-cortex using SNOWFLAKE_CORTEX_PAT env var", () =>
+    withEnv({ SNOWFLAKE_CORTEX_PAT: "test-pat" }, () =>
+      Effect.gen(function* () {
+        const plugin = yield* PluginV2.Service
+        yield* plugin.add(SnowflakeCortexPlugin)
+        const result = yield* plugin.trigger(
+          "aisdk.sdk",
+          {
+            model: model("snowflake-cortex", "claude-sonnet-4-6"),
+            package: "@ai-sdk/openai-compatible",
+            options: { name: "snowflake-cortex", baseURL: "https://test.snowflakecomputing.com/api/v2/cortex/v1" },
+          },
+          {},
+        )
+        expect(result.sdk).toBeDefined()
+      }),
+    ),
+  )
+
+  it.effect("falls back to options.apiKey when SNOWFLAKE_CORTEX_PAT env var is absent", () =>
+    withEnv({ SNOWFLAKE_CORTEX_PAT: undefined }, () =>
+      Effect.gen(function* () {
+        const plugin = yield* PluginV2.Service
+        yield* plugin.add(SnowflakeCortexPlugin)
+        const result = yield* plugin.trigger(
+          "aisdk.sdk",
+          {
+            model: model("snowflake-cortex", "claude-sonnet-4-6"),
+            package: "@ai-sdk/openai-compatible",
+            options: {
+              name: "snowflake-cortex",
+              baseURL: "https://test.snowflakecomputing.com/api/v2/cortex/v1",
+              apiKey: "options-pat",
+            },
+          },
+          {},
+        )
+        expect(result.sdk).toBeDefined()
+      }),
+    ),
+  )
+
+  it.effect("sets includeUsage on the SDK options", () =>
+    withEnv({ SNOWFLAKE_CORTEX_PAT: "test-pat" }, () =>
+      Effect.gen(function* () {
+        const plugin = yield* PluginV2.Service
+        const captured: Record<string, unknown>[] = []
+        yield* plugin.add(SnowflakeCortexPlugin)
+        yield* plugin.add({
+          id: PluginV2.ID.make("inspector"),
+          effect: Effect.succeed({
+            "aisdk.sdk": (evt) =>
+              Effect.sync(() => {
+                captured.push({ ...evt.options })
+              }),
+          }),
+        })
+        yield* plugin.trigger(
+          "aisdk.sdk",
+          {
+            model: model("snowflake-cortex", "claude-sonnet-4-6"),
+            package: "@ai-sdk/openai-compatible",
+            options: { name: "snowflake-cortex", baseURL: "https://test.snowflakecomputing.com/api/v2/cortex/v1" },
+          },
+          {},
+        )
+        expect(captured[0]?.includeUsage).toBe(true)
+      }),
+    ),
+  )
+})
+
+type FetchLike = (url: string | URL | Request, init?: RequestInit) => Promise<Response>
+
+describe("cortexFetch", () => {
+  bun_it("rewrites max_tokens to max_completion_tokens", async () => {
+    const captured: RequestInit[] = []
+    const upstream: FetchLike = async (_url, init) => {
+      captured.push(init ?? {})
+      return new Response("{}", { status: 200 })
+    }
+    await cortexFetch(upstream)("https://test", {
+      method: "POST",
+      body: JSON.stringify({ model: "claude-sonnet-4-6", max_tokens: 1024 }),
+    })
+    const body = JSON.parse(captured[0].body as string)
+    expect(body.max_completion_tokens).toBe(1024)
+    expect(body.max_tokens).toBeUndefined()
+  })
+
+  bun_it("preserves body when max_tokens is absent", async () => {
+    const captured: RequestInit[] = []
+    const upstream: FetchLike = async (_url, init) => {
+      captured.push(init ?? {})
+      return new Response("{}", { status: 200 })
+    }
+    const original = JSON.stringify({ model: "claude-sonnet-4-6", temperature: 0.7 })
+    await cortexFetch(upstream)("https://test", { method: "POST", body: original })
+    expect(captured[0].body).toBe(original)
+  })
+
+  bun_it("treats 400 'conversation complete' as a stop response", async () => {
+    const upstream: FetchLike = async () =>
+      new Response(JSON.stringify({ message: "Conversation complete" }), {
+        status: 400,
+        headers: { "content-type": "application/json" },
+      })
+    const response = await cortexFetch(upstream)("https://test", {})
+    expect(response.status).toBe(200)
+    const data = (await response.json()) as { choices: { finish_reason: string }[] }
+    expect(data.choices[0].finish_reason).toBe("stop")
+  })
+
+  bun_it("passes through other 400 errors unchanged", async () => {
+    const upstream: FetchLike = async () =>
+      new Response(JSON.stringify({ message: "Invalid model" }), {
+        status: 400,
+        headers: { "content-type": "application/json" },
+      })
+    const response = await cortexFetch(upstream)("https://test", {})
+    expect(response.status).toBe(400)
+  })
+
+  bun_it("passes through non-400 errors unchanged", async () => {
+    const upstream: FetchLike = async () => new Response("Unauthorized", { status: 401 })
+    const response = await cortexFetch(upstream)("https://test", {})
+    expect(response.status).toBe(401)
+  })
+
+  bun_it("handles invalid JSON body gracefully without throwing", async () => {
+    const captured: RequestInit[] = []
+    const upstream: FetchLike = async (_url, init) => {
+      captured.push(init ?? {})
+      return new Response("{}", { status: 200 })
+    }
+    const invalidBody = "{ not json }"
+    await cortexFetch(upstream)("https://test", { method: "POST", body: invalidBody })
+    expect(captured[0].body).toBe(invalidBody)
+  })
+
+  bun_it("rewrites role:'' to role:'assistant' in streaming SSE chunks", async () => {
+    const chunk = `data: {"choices":[{"delta":{"role":"","content":"Hi"},"index":0}]}\n\n`
+    const upstream: FetchLike = async () =>
+      new Response(new ReadableStream({ start: (ctrl) => { ctrl.enqueue(new TextEncoder().encode(chunk)); ctrl.close() } }), {
+        status: 200,
+        headers: { "content-type": "text/event-stream" },
+      })
+    const response = await cortexFetch(upstream)("https://test", {})
+    const text = await response.text()
+    expect(text).toContain('"role":"assistant"')
+    expect(text).not.toContain('"role":""')
+  })
+})