Skip to content

Mistral AI ​

Rho supports Mistral AI La Plateforme as a first-party provider. Rho talks to Mistral's Chat Completions API at https://api.mistral.ai/v1.

Provider details ​

SettingValue
Providermistral
Authmistral-api-key
Environment overrideMISTRAL_API_KEY
API basehttps://api.mistral.ai/v1
Model sourceMistral /models
Metadatamodels.dev mistral
Default modelmistral-medium-latest

Create an API key in the Mistral console, then store it with Rho. Do not put the key in config.toml.

Interactive login ​

In the TUI, run:

text
/login mistral

Rho stores the key and refreshes the model list. Select a model with /model, for example:

text
/model mistral/mistral-medium-latest

Remove the stored key with:

text
/logout mistral

Environment and automation ​

For CI or development, set MISTRAL_API_KEY. It overrides a stored key.

bash
export MISTRAL_API_KEY="<api-key>"
rho --provider mistral \
  --auth mistral-api-key \
  --model mistral/mistral-medium-latest \
  run "review this project"

Models and reasoning ​

Use /config and choose Refresh model lists to fetch the current catalog from https://api.mistral.ai/v1/models. Rho lists only models that Mistral marks as supporting chat and function calling, so embedding, OCR, and moderation models stay out of the picker.

Mistral lists most models under several ids, such as mistral-medium-latest, mistral-medium-2604, and mistral-medium-3.5. The picker shows each model once, under the id that models.dev describes. Any of its other ids still works as /model mistral/<id> and in config.toml, and Rho sends that id to Mistral unchanged, so a -latest alias keeps following Mistral's latest release.

Cost, context, and reasoning controls come from models.dev under mistral. Mistral reasoning levels depend on the model, and Rho sends only the reasoning_effort values that models.dev lists for it. For example, Mistral Small and Medium accept none and high, and Mistral Large 4 accepts low, medium, and high. For models that do not reason, Rho sends no reasoning control.

Mistral returns reasoning as thinking content chunks. Rho shows them as reasoning and sends them back to the same model on later turns, as Mistral's reasoning guide recommends.