[heydryft]

Adds Runcrate (https://runcrate.ai) as an LLM provider. Runcrate serves open-weight models behind an OpenAI-compatible Chat Completions API, so it follows the existing Together AI / DeepSeek pattern:

• lib/models.ts: runcrate client with createOpenAI and base URL https://api.runcrate.ai/v1, key from the request or RUNCRATE_API_KEY; runcrate added to the providers that use the caller's key
• lib/models.json: four models (DeepSeek V4 Flash, GLM-5.3, Kimi K2.6, Qwen3 Coder 480B)
• lib/models.test.ts: runcrate added to the host and env key tables, so the existing routing and key tests cover it
• public/thirdparty/logos/runcrate.svg for the model picker, .env.template and README

Testing

• npm test: 77 tests, 77 pass (includes sends runcrate requests to api.runcrate.ai, ignoring a client base URL and the client key tests)
• npx tsc --noEmit -p .: exit 0
• npm run lint: exit 0 (only the two existing warnings)
• Live call through the new client (resolveModel('deepseek-ai/DeepSeek-V4-Flash'), getModelClient, generateText with RUNCRATE_API_KEY set) returned "OK" with finishReason: stop.

This PR was prepared with AI assistance (Claude Code). I work on Runcrate.

[vercel]

@heydryft is attempting to deploy a commit to the E2B Team on Vercel.

A member of the Team first needs to authorize it.