feat: add Mistral as a first-class model provider

[Xavier Pestel (xavierpestel-ai)]

What

Adds Mistral (https://api.mistral.ai/v1) as a first-class model provider in the onboarding wizard and provider registry, alongside the existing OpenAI-compatible providers (Fireworks, Baseten, Nebius, NVIDIA NIM).

Users can now pick Mistral during openwiki --init, paste a MISTRAL_API_KEY, and choose from preset models (Mistral Medium, Large, Small, Codestral, Ministral 8B/3B, Mistral Nemo) or paste any Mistral model ID.

Why

Mistral is one of the few major inference providers still reached through the generic openai-compatible path (which requires the user to know and type the base URL by hand). Promoting it to a first-class provider means:

• One click in the wizard instead of openai-compatible + hand-typed https://api.mistral.ai/v1.
• Preset model list (Mistral Medium/Large/Small, Codestral, Ministral 8B/3B, Mistral Nemo) with sensible default (mistral-medium-latest), following the same convention as Fireworks / Nebius / Baseten / NVIDIA NIM.
• MISTRAL_API_KEY recognised by resolveConfiguredProvider so a bare key auto-selects the provider without needing OPENWIKI_PROVIDER=mistral.
• MISTRAL_API_KEY added to the diagnostic redaction list so it can never leak into error output.
• MISTRAL_BASE_URL follows the same override convention as FIREWORKS_BASE_URL / BASETEN_BASE_URL / NVIDIA_BASE_URL for self-hosted or proxied gateways.

The Mistral API is OpenAI-compatible, so this reuses the existing ChatOpenAI transport with a Mistral base URL. No new SDK dependency is introduced.

The README already invites this kind of contribution:

If there is an inference provider or model you would like to see added, please open a PR.

And CONTRIBUTING.md cites the exact analogue as the canonical example of a well-scoped PR:

✅ Good: "Add Fireworks to the model provider list" — the provider config, its model options, and the doc line for it.

How it was tested

• pnpm run format — clean, all files unchanged.
• pnpm run lint — clean.
• pnpm test — 138 files / 2042 tests passing (baseline was 2039; +3 new tests exercise the Mistral provider).

New test coverage mirrors the patterns used for the other OpenAI-compatible providers:

• test/config/constants.test.ts: isValidProvider("mistral"), resolveConfiguredProvider({ MISTRAL_API_KEY }), default and env-override base URLs, whitespace-only override falls back to default, getDefaultModelId("mistral").
• test/config/env-behavior.test.ts: MISTRAL_BASE_URL surfaced verbatim in diagnostics (non-secret).
• test/setup/credentials/steps.test.ts: mistral follows the standard provider → api-key → model → langsmith step flow.
• test/agent/redaction.test.ts: MISTRAL_API_KEY values are redacted in diagnostic output.

Scope

One PR, one change: adds the Mistral provider. No unrelated refactors, no dependency changes, no other providers touched.

Changeset

.changeset/add-mistral-provider.md — minor bump, feat: add Mistral as a first-class model provider.

[changeset-bot]

🦋 Changeset detected

Latest commit: 227dae9

The changes in this PR will be included in the next version bump.

This PR includes changesets to release 1 package

Name	Type
openwiki	Minor

Not sure what this means? Click here to learn what changesets are.

Click here if you're a maintainer who wants to add another changeset to this PR