🐛 Patch Changes
- 2d7541b: Resolve model parameter specs through the providerless modelparams.dev by-model endpoint so gateway routes such as GitHub Copilot can expose the underlying model's thinking and reasoning controls.
- 5006434: Speed up the dashboard and lower proxy overhead: unbuffered SSE streaming, smaller render-blocking CSS, lazy-loaded modals, cached per-agent model lookups, and new indexes for hot queries.
- aad3033: Fix native
/v1/responsesforwarding so typed non-message input items such asreasoninganditem_referenceare preserved without arole, preventing ChatGPT subscription Codex backend 400 errors on multi-turn Codex requests.