the thinking mode can't be selected currently. it must be possible to select a supported thinking mode for each model in the chat UI, so that the user can decide if a model should do deep reasoning or just answer quickly.
the thinking mode can't be selected currently. it must be possible to select a supported thinking mode for each model in the chat UI, so that the user can decide if a model should do deep reasoning or just answer quickly.
hugo
added the bug label 2026-08-01 07:34:01 +00:00
Implemented in c0e4ec1. The chat composer now has a thinking-mode selector with DS4 normal Thinking preselected by default, Direct for fast answers without hidden reasoning, and Think Max when the configured context meets DS4s 393216-token requirement. The selected value is persisted, synchronized with the live runtime configuration, passed through TurnSettings to model prompt rendering and generation, and locked while any foreground or background generation is active so a tool continuation cannot change mode mid-turn. DeepSeek and GLM prompt/stop behavior remains aligned with ../ds4. Added cross-model mode propagation and effective-mode tests, updated the user guide, and visually verified the signed app by switching Direct and restoring Thinking. Full gates passed: cargo fmt --all -- --check, cargo clippy --all-targets --all-features -- -D warnings, make bundle, and cargo test --all-features (162 passed, 13 environment-dependent ignored).
Implemented in c0e4ec1. The chat composer now has a thinking-mode selector with DS4 normal Thinking preselected by default, Direct for fast answers without hidden reasoning, and Think Max when the configured context meets DS4s 393216-token requirement. The selected value is persisted, synchronized with the live runtime configuration, passed through TurnSettings to model prompt rendering and generation, and locked while any foreground or background generation is active so a tool continuation cannot change mode mid-turn. DeepSeek and GLM prompt/stop behavior remains aligned with ../ds4. Added cross-model mode propagation and effective-mode tests, updated the user guide, and visually verified the signed app by switching Direct and restoring Thinking. Full gates passed: cargo fmt --all -- --check, cargo clippy --all-targets --all-features -- -D warnings, make bundle, and cargo test --all-features (162 passed, 13 environment-dependent ignored).
Blocking a user prevents them from interacting with repositories, such as opening or commenting on pull requests or issues. Learn more about blocking a user.
the thinking mode can't be selected currently. it must be possible to select a supported thinking mode for each model in the chat UI, so that the user can decide if a model should do deep reasoning or just answer quickly.
Implemented in
c0e4ec1. The chat composer now has a thinking-mode selector with DS4 normal Thinking preselected by default, Direct for fast answers without hidden reasoning, and Think Max when the configured context meets DS4s 393216-token requirement. The selected value is persisted, synchronized with the live runtime configuration, passed through TurnSettings to model prompt rendering and generation, and locked while any foreground or background generation is active so a tool continuation cannot change mode mid-turn. DeepSeek and GLM prompt/stop behavior remains aligned with ../ds4. Added cross-model mode propagation and effective-mode tests, updated the user guide, and visually verified the signed app by switching Direct and restoring Thinking. Full gates passed: cargo fmt --all -- --check, cargo clippy --all-targets --all-features -- -D warnings, make bundle, and cargo test --all-features (162 passed, 13 environment-dependent ignored).