Roster for Mistral

Use Mistral as the resolver model behind Roster.

Configure a Mistral model to interpret workflow questions and help Roster select the right participants for approval, ownership, review, escalation, and human handoff.

Roster supplies the governed organizational context. Mistral interprets the request through its Chat Completions API.

Configure MistralReview supported models

Last verified 07/13/2026

Mistral provides interpretation. Roster provides organizational control.

Roster maintains:

Who should approve the Atlas infrastructure change for the European region?

Roster validates and expands the model's participant selection before returning the result.

  • Projects
  • Participants
  • Users and groups
  • Directory membership
  • Labels and metadata
  • Delegations
  • Access rules
  • Resolution history

Mistral interprets the natural-language request. Roster remains authoritative for participants, authorization, and audit.

How the integration works

Roster sends the resolution task to Mistral's Chat Completions API and validates the structured selection against its own participant and authorization context.

Agent, workflow, or application
   ↓
Roster Resolve
   ↓
Authorized participant context
   ↓
Mistral Chat Completions API
   ↓
Structured participant selection
   ↓
Active users, groups, roles, or delegates

Why use Mistral with Roster?

Direct provider integration
Roster connects directly to Mistral's public API without requiring an intermediary gateway.
Structured outputs
Mistral supports structured response formats through Chat Completions for compatible models.
Provider choice
Organizations can use Mistral as part of a broader model-vendor, cost, geography, or procurement strategy.
OpenAI-like request conventions
Mistral's Chat Completions interface follows familiar message-based API conventions, simplifying operational adoption for many platform teams.

Configure Mistral

Roster defaults to ROSTER_MODEL_API=chat-completions for direct Mistral configurations.

MISTRAL_API_KEY=<mistral-api-key>
MISTRAL_BASE_URL=https://api.mistral.ai/v1
ROSTER_MODEL_PROVIDER=mistral
ROSTER_MODEL_NAME=mistral-large-2512

MISTRAL_BASE_URL is optional when using Mistral's standard public endpoint. Store the API key in a secret manager or protected runtime environment.

Recommended model — Roster's current Full Resolve recommendation is mistral-large-2512. Roster recommends the larger model over smaller Mistral options for complex prompts involving several responsibilities, constraints, or possible participants. Other Mistral models may be useful for cost or latency optimization, but they should remain test or canary candidates until they pass the full Resolve evaluation suite.

Reasoning-effort behavior — direct Mistral requests currently ignore ROSTER_MODEL_EFFORT even though Roster accepts the variable globally. Validate quality and performance through model selection, participant-model design, metadata quality, output limits, representative queries, and full Resolve evaluations.

Example resolution

An infrastructure workflow asks:

Who should review and approve this production network change for Atlas Europe?
  1. Roster identifies the authorized Atlas project.
  2. Roster supplies the relevant participant definitions and metadata.
  3. Roster includes materialized membership and active delegations.
  4. Roster sends the interpretation task to the configured Mistral model.
  5. Roster validates the returned selection.
  6. Roster resolves the participant into active users, groups, or delegated contacts.
  7. Roster records the Resolve request and model run.

The workflow remains responsible for delivering and enforcing the review.

Structured output and observability

Mistral supports structured outputs for compatible Chat Completions models, but the exact model/API combination must be tested. The selected model needs to handle required JSON fields, participant identifiers, multiple matching responsibilities, no-match results, ambiguity, delegation context, and repeated equivalent queries.

Do not promote a model based solely on successful plain-text completions.

Roster Model Runs records provider, model, API type, status, live or test mode, latency, token usage, estimated cost, request and trace identifiers, input and output (subject to PII settings), and error details. Use these records to compare Mistral latency, cost, error rate, and resolution quality against the same evaluation set used for other providers.

Production checklist

Verify API-key access
Confirm the credential works against the intended endpoint.
Pin the intended model ID
Avoid silent upgrades between versions by pinning the exact model name.
Validate structured output
Confirm the model produces parseable, schema-conformant responses across representative queries.
Run Resolve evaluations
Include ambiguous and no-result cases, participant memberships, and delegations.
Measure latency and cost
Compare against other approved providers before promoting to production.
Review provider policy
Review Mistral retention, processing, regional requirements, and rate limits.
Define error handling
Specify retry, fallback, or manual-routing behavior for provider errors.

Frequently asked questions

Roster uses Mistral's Chat Completions API.

Add Mistral interpretation to governed workflow routing

Use Mistral to interpret the question while Roster controls participant data, project scope, directory membership, delegation, authorization, and observability.

Configure MistralReview model-provider documentationExplore all integrations