feat(agent): M4b — OpenAI-compatible HTTP provider (Qwen/Kimi) #11
Loading…
Add table
Add a link
Reference in a new issue
No description provided.
Delete branch "feat/m4b-http-provider"
Deleting a branch is permanent. Although the deleted branch may continue to exist for a short time before it actually gets removed, it CANNOT be undone in most cases. Continue?
Roadmap M4b: the real LLM provider.
OpenAiCompatimplementsLlmProviderover reqwest/rustls — one adapter for Qwen3.6/Ollama and Kimi K2/UmansAI (FR-AGT-04/05), backend-config-driven.Changes
OpenAiCompat::new(base, api_key, model)→ POST{base}/chat/completionswith messages + tools, optional bearer.build_body(role mapping, tool specs, tool_choice) andparse_turn(tool_calls with JSON-string arguments → Value, final text, garbage→error). No live LLM or mock server in CI.Verification
just qagreen; agent 9 tests (5 loop + 4 mapping).Next: M5 integrations (transport/lodging/activities) then M6 — wire agent + tools into the trip builder, replacing StubBuilder. Traces: FR-AGT-04/05 · tech-spec 04 §4.1.
- OpenAiCompat implements LlmProvider over reqwest (rustls): POST {base}/chat/completions with messages + tools, bearer auth optional. One adapter for both Qwen3.6/Ollama and Kimi K2/UmansAI (FR-AGT-04/05). - request/response mapping is pure (build_body / parse_turn) and unit-tested — role mapping, tool specs, tool_call parsing (arguments JSON-string → Value), final-text, garbage-errors. No live LLM / mock server needed in CI. just qa green; agent 9 tests (5 loop + 4 mapping). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>