A reverse-engineered Grok gateway that speaks OpenAI

What it does

Grok2API is a FastAPI-based gateway that reverse-engineers Grok’s web capabilities and exposes them through OpenAI-compatible endpoints — chat completions, image generation, video creation, and even Anthropic-style messages. It also bundles a web admin panel, chat interface, and image gallery so you aren’t stuck in curl land.

The interesting bit

The project treats Grok’s web tier as an opaque backend and builds a full translation layer on top: account pooling with tier-aware routing, local media caching, token usage estimation, and configurable proxy chains for egress and Cloudflare clearance. That’s a lot of operational sophistication for what is essentially a protocol adapter.

Key highlights

• OpenAI and Anthropic API compatibility, including streaming, reasoning output, and function tool passthrough
• Multi-account pool with automatic failover, tier selection (basic/super/heavy), and usage synchronization
• Supports text-to-image, image editing, text-to-video, and image-to-video through unified or dedicated endpoints
• Local media caching with configurable URL proxying to avoid 403s on Grok-hosted assets
• Built-in WebUI with admin controls, chat, masonry image grid, and voice chat via ChatKit
• Deployable via Docker Compose, Vercel, or Render

Caveats

• Explicitly labeled “for learning and research exchange only”; using it violates Grok’s terms of service
• Requires ongoing reverse-engineering maintenance as Grok’s web interface changes
• app.app_url must be configured or media links break with 403 errors

Verdict

Worth a look if you need Grok’s capabilities in a pipeline that expects standard LLM APIs and you’re comfortable with the maintenance burden and legal gray area. Skip it if you want a supported, stable provider contract — this is a cat-and-mouse game by design.