Add Mistral AI provider

[Genmin]

Summary

• Adds a first-class Mistral AI provider backed by https://api.mistral.ai/v1/chat/completions
• Registers Mistral across core provider resolution, smart routing, fallback/race helpers, auxiliary calls, CLI /provider add, API onboarding, provider validation, and streaming endpoint maps
• Adds MISTRAL_API_KEY to .env.example and a disabled default mistral config entry using mistral-large-latest

Notes

• The provider mirrors the existing OpenAI-compatible Groq/BOA shape and supports non-streaming, streaming, and tool-call requests.
• Response content extraction handles both plain string content and chunk-array content.

Validation

• npx tsc --noEmit --target ES2020 --module commonjs --moduleResolution node --esModuleInterop --allowSyntheticDefaultImports --skipLibCheck --strict false --lib ES2020,DOM providers/mistral.ts
• npx esbuild providers/mistral.ts --bundle --platform=node --target=node18 --outfile=/tmp/aiden-mistral-provider.js
• git diff --check

Existing local build blockers observed

• npm ci currently fails before installing because package.json and package-lock.json are out of sync (aiden-os@3.17.0 missing from the lockfile).
• After installing without modifying the lockfile, npm run build stops on an existing minimatch import/type issue in core/toolRegistry.ts.
• npm run build:api also stops on unresolved @aws-sdk/client-s3 from unzipper in this local install.

Closes #25

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 0dc40ab3e4

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[chatgpt-codex-connector]

Normalize Mistral content before returning from callLLM

Adding mistral to OPENAI_COMPAT_ENDPOINTS routes it through the generic OpenAI branch in callLLM, which returns d?.choices?.[0]?.message?.content verbatim; Mistral can return that field as an array of content parts (as already handled in providers/mistral.ts). In that case downstream planner/retry paths that do raw.trim() will hit raw.trim is not a function, causing failed turns and unintended fallback behavior whenever Mistral is selected.

Useful? React with 👍 / 👎.

[chatgpt-codex-connector]

Parse Mistral content parts in provider-racing helper

By adding mistral to COMPAT_ENDPOINTS, fetchProviderResponse now returns Mistral responses through the same unnormalized message.content path used for string-only providers. If Mistral returns content-part arrays, raceProviders calls winner.text.trim()/result.text.trim() and treats the call as failed, so pinned or raced Mistral providers can be skipped even when the API returned a valid answer.

Useful? React with 👍 / 👎.