You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
fix(opencodego): reasoning fallback + simplify OpenCode Go catalog (#33)
* refactor: remove opencodego client and legacy model tier configs
Drop the OpenCode Go provider and hawk ModelKey tier tables in favor of
live catalog data and deployment routing.
* feat(opencodego): OpenAI-compatible client with live model discovery
Reuse chat/completions and GET /models like other gateways: thin
OpenCodeGoClient wrapper, dynamic catalog fetch, and filter out models
that require the messages API (MiniMax, Qwen3.x).
* feat(opencodego): dual-protocol routing with shared DualProtocolPair
Route MiniMax/Qwen via /v1/messages with chat/completions fallback, discover all models live, and centralize OpenAI/Anthropic switching for OpenCode Go and MiMo.
* refactor(client): rename DualProtocolPair to ProtocolRouter
Reuse OpenAIClient and AnthropicClient directly for chat-first OpenCode Go models; keep ProtocolRouter for shared fallback orchestration only.
* fix(opencodego): recover reasoning-only MiniMax streams via chat fallback
When /v1/messages streams thinking without answer text, retry through
non-streaming chat/completions before streaming fallback so MiniMax M3
returns usable replies on OpenCode Go.
* fix(ci): satisfy lint after main merge
Remove redundant nil checks in ProtocolRouter and delete unused tier helpers.
* fix(ci): remove unused contains helper
0 commit comments