[XYAIStudio]

Summary

Models page local list (Ollama / registered GGUF) now supports two desktop-friendly actions:

1. 测速 (speed test) — pings the local runtime with a tiny fixed prompt and shows total latency, time-to-first-token, and tokens/sec.
2. 设为默认 (set as default) — marks a registered local model as the chat default so new conversations and the model picker prefer it.

How default is stored

• Per-user preferred_model in user preferences (provider/model, e.g. Ollama (Local)/tiny). The chat composer already initializes new threads from this value.
• Global active_model in settings KV (same provider/model ref). AUTO expert routing and the Models pool fall back to this when the user has no preferred model.
• Clear default (取消默认) clears the preferred model and, when it matches, the global active model.
• Set-default is allowed when the tag is registered or already enabled on the local provider, even if Ollama is currently down.

How speed test is measured

• POST /api/local-models/speed-test with { name }.
• First tries native Ollama POST {root}/api/generate (streamed) with Reply with the single word ok. and num_predict=16.
• Falls back to OpenAI-compatible {base}/v1/chat/completions if generate returns 404/405.
• latency_ms: wall time of the request.
• ttft_ms: time until the first non-empty token.
• tokens_per_sec: Ollama eval_count / eval_duration when present, otherwise tokens / elapsed.
• Timeout 45s on both server and UI; the UI can cancel without blocking the page. Last result is kept in sessionStorage on the model row.
• If Ollama is down, 测速 is disabled with an explanation; 设为默认 stays available for registered models.

Target branch

• Base is main (requested)

Type of change

• New feature

Test plan

• uv run pytest tests/unit/agents/test_local_speed.py tests/unit/agents/test_local_default.py tests/integration/test_local_models_default.py
• npx vitest run for LocalHardwarePanel + localSpeedResults (7 passed)
• uv run ruff check + mypy --strict src/octop
• cd dashboard && npx tsc --noEmit
• make test — 3497 passed, 15 skipped
• GitHub CI — 5 checks green on 971e5d75
• Live Ollama UI click-through (not available in this environment)

Checklist

• Updated CHANGELOG.md
• Updated docs/api.md