v1.1.0

v1.1.0 — Qwen Responses API Migration

Highlights

• Qwen now uses the Responses API (/v1/responses) instead of Chat Completions. This enables reasoning.effort for granular reasoning control (replacing the deprecated enable_thinking), server-side session caching via x-dashscope-session-cache header, and lays groundwork for previous_response_id multi-turn conversations.
• DeepSeek remains on Chat Completions — both vendors share the same Provider interface, transparent to callers.

Changes

New

• responsesProvider — universal Responses API provider with full status handling (completed, incomplete, failed, cancelled)
• qwenResponsesDialect — maps ThinkingSpec to reasoning.effort (max/none), injects session cache header
• Debug-level logging of reasoning item count and summary length for Responses API parity with Chat Completions

Modified

• NewProvider routes PROVIDER=qwen to responsesProvider (was openaiProvider)
• Rate limit handling and retry mechanism improvements

Removed

• qwenDialect and qwenThinkingEnabled (Chat Completions Qwen dialect, superseded)

Upgrade Notes

No configuration changes required. Existing .env with PROVIDER=qwen works out of the box — the server automatically selects the Responses API based on the vendor setting.

Verification

• Build, test, vet, format, and lint all pass
• Local HTTP smoke test confirmed: reasoning tokens reported, proper usage accounting