mirror of
https://github.com/ggozad/oterm.git
synced 2026-10-09 16:53:20 +02:00
781bc7c043The window only reached the summarize capability after the first reply reported usage, so a reopened long chat on a local server sent its whole history uncompacted: Ollama silently drops the oldest context and vLLM rejects the request. The chat now asks the server on mount, after each reply whether or not it reported usage, and again after the chat is edited, which previously kept the old model's window. The system prompt check that drives the summary notification stays, with the invariant it relies on stated: oterm sends its system prompt as instructions and none of its capabilities add system prompts. The turn helper in the widget tests waited for exactly two messages, which a preloaded chat can never reach; it now waits for two more than were there. |
||
|---|---|---|
| .. | ||
| __init__.py | ||
| test_caps.py | ||
| test_chat_container.py | ||
| test_flexible_input.py | ||
| test_image_widget.py | ||
| test_model_select.py | ||
| test_prompt_history.py | ||
| test_tool_select.py | ||