Switch Models Per Task

How assigned vs default model selection works, and what suffix matching does to your ids.

Per mission, set the assigned model when creating/updating it (assigned_model parameter, app/services/mission_service.py:239). Without one, your default model resolution applies.
Flowmanner matches keys tolerantly: if a BYOK key declares deepseek-chat/deepseek-v4-flash and you request a suffix match (deepseek-v4-flash), the key still wins the route (llm_router.py:858-870).
The wire spelling follows the key: when routing matched only by suffix, Flowmanner sends the key's declared spelling upstream — strict gateways reject spellings they did not declare (llm_router.py:512-524). You keep seeing your original id in results.
Reasoning-style models spend part of their window on internal reasoning; size budgets accordingly (Models).

Last updated 2026-08-25 (git-derived)

How assigned vs default model selection works, and what suffix matching does to your ids.

Per mission, set the assigned model when creating/updating it (assigned_model parameter, app/services/mission_service.py:239). Without one, your default model resolution applies.
Flowmanner matches keys tolerantly: if a BYOK key declares deepseek-chat/deepseek-v4-flash and you request a suffix match (deepseek-v4-flash), the key still wins the route (llm_router.py:858-870).
The wire spelling follows the key: when routing matched only by suffix, Flowmanner sends the key's declared spelling upstream — strict gateways reject spellings they did not declare (llm_router.py:512-524). You keep seeing your original id in results.
Reasoning-style models spend part of their window on internal reasoning; size budgets accordingly (Models).

Last updated 2026-08-25 (git-derived)