0.11.0

0.11.0

Reasoning effort, made honest and made settable.

Added

• What "Auto" reasoning sends, per provider, in Settings → Models. Each provider card whose family takes a default now has an Auto reasoning control: provider default, off, or a level the family can express. It writes the family's own spelling — Anthropic's effort and thinking: 'off', OpenAI's reasoningEffort, DeepSeek's and Kimi's thinking — through a route that changes those keys and nothing else, because the setting lives beside the API key. The VS Code panel's Providers tab shows the value in effect.

Fixed

• The effort button kept saying the rung you picked on another model. Pick xhigh on Claude, switch the session to Kimi K3, and the request went out with high while the button still read xhigh. Both the web composer and the VS Code panel now show the rung that will actually be sent, and explain why when it differs.

Verified

Harness 2,512 / 0 · web unit 202 / 0 · web layout 52 / 0 · the control driven in a real browser and read back from the server · panel tunnel and VS Code panel probes in a real editor.

Install

npx github:suhail-akhtar/aico#v0.11.0 serve
npm install -g github:suhail-akhtar/aico#v0.11.0

The VS Code extension aico-vscode-0.6.4.vsix is attached; install it with Extensions: Install from VSIX….