fix(runtime): restore last-used reasoning effort on model switch by Tryanks · Pull Request #283 · Tryanks/tcode

Tryanks · 2026-08-26T05:23:00Z

Problem

Switching the active model (draft or established session) unconditionally cleared option_selections, so reasoning effort always reset to the model's default — e.g. every new conversation switched to gpt-5.6-sol came up at low.

Fix

Each session's meta already persists the effort it ran at, so no new storage is needed. set_active_model now restores the reasoningEffort last used with exactly that (provider, model) from session history (most recently updated match, archived included) after clearing stale selections. Keyed per model, the remembered value is always one the target model accepts.

sessions.rs: new remembered_effort() helper next to draft_defaults
options.rs: all three clear sites in set_active_model re-extend with the remembered effort

Test

model_switch_restores_last_effort_used_with_that_model: last gpt-5.6-sol session ran at max, newest session is claude-fable-5; switching a fresh draft back to sol restores max. Full cargo test -p tcode-runtime: 124 passed.

Switching the active model cleared option_selections unconditionally, so the effort always reset to the model's default (low for gpt-5.6-sol). Restore the reasoningEffort last used with exactly that (provider, model) from session history instead — no new storage, and the remembered value is always one the model accepts.

Tryanks · 2026-08-26T05:23:00Z

Problem

Switching the active model (draft or established session) unconditionally cleared option_selections, so reasoning effort always reset to the model's default — e.g. every new conversation switched to gpt-5.6-sol came up at low.

Fix

Each session's meta already persists the effort it ran at, so no new storage is needed. set_active_model now restores the reasoningEffort last used with exactly that (provider, model) from session history (most recently updated match, archived included) after clearing stale selections. Keyed per model, the remembered value is always one the target model accepts.

sessions.rs: new remembered_effort() helper next to draft_defaults
options.rs: all three clear sites in set_active_model re-extend with the remembered effort

Test

model_switch_restores_last_effort_used_with_that_model: last gpt-5.6-sol session ran at max, newest session is claude-fable-5; switching a fresh draft back to sol restores max. Full cargo test -p tcode-runtime: 124 passed.

Switching the active model cleared option_selections unconditionally, so the effort always reset to the model's default (low for gpt-5.6-sol). Restore the reasoningEffort last used with exactly that (provider, model) from session history instead — no new storage, and the remembered value is always one the model accepts.

Tryanks · 2026-08-26T05:23:00Z

Problem

Switching the active model (draft or established session) unconditionally cleared option_selections, so reasoning effort always reset to the model's default — e.g. every new conversation switched to gpt-5.6-sol came up at low.

Fix

Each session's meta already persists the effort it ran at, so no new storage is needed. set_active_model now restores the reasoningEffort last used with exactly that (provider, model) from session history (most recently updated match, archived included) after clearing stale selections. Keyed per model, the remembered value is always one the target model accepts.

sessions.rs: new remembered_effort() helper next to draft_defaults
options.rs: all three clear sites in set_active_model re-extend with the remembered effort

Test

model_switch_restores_last_effort_used_with_that_model: last gpt-5.6-sol session ran at max, newest session is claude-fable-5; switching a fresh draft back to sol restores max. Full cargo test -p tcode-runtime: 124 passed.

Switching the active model cleared option_selections unconditionally, so the effort always reset to the model's default (low for gpt-5.6-sol). Restore the reasoningEffort last used with exactly that (provider, model) from session history instead — no new storage, and the remembered value is always one the model accepts.

Tryanks · 2026-08-26T05:23:00Z

Problem

Switching the active model (draft or established session) unconditionally cleared option_selections, so reasoning effort always reset to the model's default — e.g. every new conversation switched to gpt-5.6-sol came up at low.

Fix

Each session's meta already persists the effort it ran at, so no new storage is needed. set_active_model now restores the reasoningEffort last used with exactly that (provider, model) from session history (most recently updated match, archived included) after clearing stale selections. Keyed per model, the remembered value is always one the target model accepts.

sessions.rs: new remembered_effort() helper next to draft_defaults
options.rs: all three clear sites in set_active_model re-extend with the remembered effort

Test

model_switch_restores_last_effort_used_with_that_model: last gpt-5.6-sol session ran at max, newest session is claude-fable-5; switching a fresh draft back to sol restores max. Full cargo test -p tcode-runtime: 124 passed.

Switching the active model cleared option_selections unconditionally, so the effort always reset to the model's default (low for gpt-5.6-sol). Restore the reasoningEffort last used with exactly that (provider, model) from session history instead — no new storage, and the remembered value is always one the model accepts.

Tryanks · 2026-08-26T05:23:00Z

Problem

Switching the active model (draft or established session) unconditionally cleared option_selections, so reasoning effort always reset to the model's default — e.g. every new conversation switched to gpt-5.6-sol came up at low.

Fix

Each session's meta already persists the effort it ran at, so no new storage is needed. set_active_model now restores the reasoningEffort last used with exactly that (provider, model) from session history (most recently updated match, archived included) after clearing stale selections. Keyed per model, the remembered value is always one the target model accepts.

sessions.rs: new remembered_effort() helper next to draft_defaults
options.rs: all three clear sites in set_active_model re-extend with the remembered effort

Test

model_switch_restores_last_effort_used_with_that_model: last gpt-5.6-sol session ran at max, newest session is claude-fable-5; switching a fresh draft back to sol restores max. Full cargo test -p tcode-runtime: 124 passed.

Switching the active model cleared option_selections unconditionally, so the effort always reset to the model's default (low for gpt-5.6-sol). Restore the reasoningEffort last used with exactly that (provider, model) from session history instead — no new storage, and the remembered value is always one the model accepts.

Tryanks · 2026-08-26T05:23:00Z

Problem

Switching the active model (draft or established session) unconditionally cleared option_selections, so reasoning effort always reset to the model's default — e.g. every new conversation switched to gpt-5.6-sol came up at low.

Fix

Each session's meta already persists the effort it ran at, so no new storage is needed. set_active_model now restores the reasoningEffort last used with exactly that (provider, model) from session history (most recently updated match, archived included) after clearing stale selections. Keyed per model, the remembered value is always one the target model accepts.

sessions.rs: new remembered_effort() helper next to draft_defaults
options.rs: all three clear sites in set_active_model re-extend with the remembered effort

Test

model_switch_restores_last_effort_used_with_that_model: last gpt-5.6-sol session ran at max, newest session is claude-fable-5; switching a fresh draft back to sol restores max. Full cargo test -p tcode-runtime: 124 passed.

Switching the active model cleared option_selections unconditionally, so the effort always reset to the model's default (low for gpt-5.6-sol). Restore the reasoningEffort last used with exactly that (provider, model) from session history instead — no new storage, and the remembered value is always one the model accepts.

Tryanks · 2026-08-26T05:23:00Z

Problem

Switching the active model (draft or established session) unconditionally cleared option_selections, so reasoning effort always reset to the model's default — e.g. every new conversation switched to gpt-5.6-sol came up at low.

Fix

Each session's meta already persists the effort it ran at, so no new storage is needed. set_active_model now restores the reasoningEffort last used with exactly that (provider, model) from session history (most recently updated match, archived included) after clearing stale selections. Keyed per model, the remembered value is always one the target model accepts.

sessions.rs: new remembered_effort() helper next to draft_defaults
options.rs: all three clear sites in set_active_model re-extend with the remembered effort

Test

model_switch_restores_last_effort_used_with_that_model: last gpt-5.6-sol session ran at max, newest session is claude-fable-5; switching a fresh draft back to sol restores max. Full cargo test -p tcode-runtime: 124 passed.

Switching the active model cleared option_selections unconditionally, so the effort always reset to the model's default (low for gpt-5.6-sol). Restore the reasoningEffort last used with exactly that (provider, model) from session history instead — no new storage, and the remembered value is always one the model accepts.

Tryanks · 2026-08-26T05:23:00Z

Problem

Switching the active model (draft or established session) unconditionally cleared option_selections, so reasoning effort always reset to the model's default — e.g. every new conversation switched to gpt-5.6-sol came up at low.

Fix

Each session's meta already persists the effort it ran at, so no new storage is needed. set_active_model now restores the reasoningEffort last used with exactly that (provider, model) from session history (most recently updated match, archived included) after clearing stale selections. Keyed per model, the remembered value is always one the target model accepts.

sessions.rs: new remembered_effort() helper next to draft_defaults
options.rs: all three clear sites in set_active_model re-extend with the remembered effort

Test

model_switch_restores_last_effort_used_with_that_model: last gpt-5.6-sol session ran at max, newest session is claude-fable-5; switching a fresh draft back to sol restores max. Full cargo test -p tcode-runtime: 124 passed.

Switching the active model cleared option_selections unconditionally, so the effort always reset to the model's default (low for gpt-5.6-sol). Restore the reasoningEffort last used with exactly that (provider, model) from session history instead — no new storage, and the remembered value is always one the model accepts.

Tryanks merged commit eaeac60 into mainAug 26, 2026
3 checks passed

Tryanks deleted the fix/remember-model-effort branch August 26, 2026 05:29