The brain behind the robot.

One API to route every robot task to the right open foundation model — GR00T, OpenVLA, Fable, o-series, Gemini Robotics. Retarget any skill to any embodiment. Every decision signed and chain-anchored for forensic-grade audit.

$ curl https://api.cortex.brain/v1/orchestrate \
    -H 'x-cortex-api-key: ck_demo_...' \
    -d '{"task":"pick up the red cup","robot_id":"robot_g1_demo"}'

→ model: nvidia/gr00t-n1   confidence: 0.95
→ plan: 6 steps · 5.0s total
→ signature: ed25519:7d3f...   prev_hash: 0a91...

Orchestrator

One POST. We route your task to the best open robot foundation model — GR00T, OpenVLA, Fable, o-series, Gemini Robotics — using task-aware scoring. The plan comes back ready to execute.

Cross-embodiment translator

Train once on G1 (29 DoF), deploy on H1 (37 DoF) or any URDF you register. Joint-graph-aware retargeting with a mapping table you can audit, frame-by-frame.

Forensic provenance

Every decision, every retarget, hashed and signed with ed25519, chained per-customer. When the insurance adjuster comes, you hand them a verifiable record.

Built on the same stack as Cortex Studio.

Next.js 15, TypeScript, Tailwind v4, Drizzle, @noble/ed25519. Routing via OpenRouter, xAI, and DapiCloud (Anthropic). Deploys to Vercel for the web tier and Hetzner cerebro-build for heavy GPU services.