shumengya/sproutclaw
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

fix(ai): handle explicit thinking off across providers closes #2490

Browse Source
This commit is contained in:
Mario Zechner
2026-03-22 20:27:44 +01:00
parent 6129971c04
commit d1613e3f53
8 changed files with 246 additions and 28 deletions
Download Patch File Download Diff File Expand all files Collapse all files

1
packages/ai/CHANGELOG.md
View File

@@ -11,6 +11,7 @@
- Fixed OpenAI Responses replay for foreign tool-call item IDs by hashing foreign `function_call.id` values into bounded `fc_<hash>` IDs instead of preserving backend-specific normalized shapes that OpenAI Codex rejects.
- Fixed Anthropic thinking disable handling to send `thinking: { type: "disabled" }` for reasoning-capable models when thinking is explicitly off, and added payload and env-gated end-to-end coverage for the Anthropic provider ([#2022](https://github.com/badlogic/pi-mono/issues/2022))
- Fixed explicit thinking disable handling across Google, Google Vertex, Gemini CLI, OpenAI Responses, Azure OpenAI Responses, and OpenRouter-backed OpenAI-compatible completions. Gemini 3 models now fall back to the lowest supported thinking level when full disable is not supported, and OpenAI/OpenRouter reasoning models now send explicit `none` effort instead of relying on provider defaults ([#2490](https://github.com/badlogic/pi-mono/issues/2490))
## [0.61.1] - 2026-03-20