Back to rankings

Mistral Vibe: Programmatic Mode Gains --model and --system-prompt-file Flags

Mistral VibeView original changelog

Mistral Vibe 3.0.4 added two flags for programmatic mode. vibe -p --model ALIAS runs a single invocation with a chosen model without changing the default, and vibe -p --system-prompt-file PATH replaces Vibe's own instructions in the system prompt with the file's text. A new pair of settings also caps output tokens for agent and compaction requests.

Key Takeaways

  • vibe -p --model ALIAS runs one invocation with a chosen model without altering the default.
  • --system-prompt-file PATH replaces Vibe's own instructions in the system prompt.
  • agent_max_output_tokens and compaction_max_output_tokens cap output on any provider.
  • A cap of 0 (default) sends a limit only where the provider requires one.
  • Setup now warns when a shell API key overrides the saved key.
  • /whoami shows whether credentials come from the shell, .env or keyring.

Pick a Model Per Run

vibe -p --model ALIAS runs a programmatic invocation with the given model without changing the user's default model. Scripts and CI jobs can now choose a model per call instead of editing configuration.

Replace the System Prompt

vibe -p --system-prompt-file PATH replaces Vibe's own instructions in the system prompt with the text of the given file. This gives users full control over agent instructions for specialized automation.

Output Token Caps

Two new settings, agent_max_output_tokens and compaction_max_output_tokens, cap the output of every agent and compaction request on any provider. The cap is held to what the model and its context window leave room for. At 0, the default, a cap is still sent only where the provider requires one.

Smaller Setup Improvements

Setup now warns when an API key set in the shell overrides the key it saves, and /whoami shows which credential source (shell, .env or keyring) is in use. The provider option preserve_api_base = true keeps a custom api_base that /whoami would otherwise overwrite, which helps split-horizon or proxied deployments.