The operation layer, over HTTP.

One call turns "there is a microwave in front of me" into a grounded, safety-gated, machine-executable operation package: every control with a visual anchor, ordered procedures with per-step verification signals, force limits, hazards and interlocks — re-expressed from the appliance's own documentation, never guessed by a model at runtime.

Instant start — no signup

One command mints a working key with 2 operation packages included:

KEY=$(curl -s -X POST https://api.operandi.cc/v1/trial | python3 -c "import json,sys;print(json.load(sys.stdin)['api_key'])")
curl -s https://api.operandi.cc/v1/operate/coffee-maker-bialetti-moka-express \
  -H "Authorization: Bearer $KEY"

Agents: pip install operandi-mcp does this for you automatically — zero config, answering in seconds.

Base URL & accounts

https://api.operandi.cc

When your trial is spent, sign up (free tier: 10 packages/month) and mint a permanent key (shown once, prefix ok_live_):

curl -X POST https://api.operandi.cc/v1/auth/signup \
  -H "Content-Type: application/json" \
  -d '{"email":"[email protected]","password":"********"}'

# then create a key with your session: POST /v1/auth/keys

Send the key on every call — either header works:

Authorization: Bearer ok_live_...        # preferred
X-API-Key: ok_live_...                   # equivalent

Free tier: 10 operation packages per month across all delivery routes, plus unmetered catalog/browse calls. Past the cap you get 402 with an X-Upgrade-Url header — nothing breaks silently.

The two calls that matter

1 · Identify what you're looking at

curl -s https://api.operandi.cc/v1/identify \
  -H "Authorization: Bearer $KEY" -H "Content-Type: application/json" \
  -d '{"text":"Samsung MS23K3513 microwave nameplate"}'

Send whatever text your perception produced — nameplate OCR, model strings, a human hint. Returns ranked appliance matches with slugs and package availability.

2 · Pull the operation package

curl -s https://api.operandi.cc/v1/operate/samsung-ms23k3513 \
  -H "Authorization: Bearer $KEY"

Returns the full robot-ready package in one response: identity · controls (with visual_anchor, interaction physics: motion, force class, travel, expected feedback) · ordered procedures with per-step verify signals · safety envelope (hazards, interlocks, never-do) · the executable ROP behaviour tree · trust tier and validation history.

All endpoints

What makes the data trustworthy

• Grounded: a deterministic checker verifies every number a package claims exists in its source document. Current corpus: ~96% verified-clean; flagged packages say so.
• Honest tiers: every ROP carries a validation_tier — auto → sim_validated (kinematic walk passes) → phys_validated (software contact-physics screen) → robot_cleared (human sign-off; hardware). Nothing is over-claimed: if a package hasn't passed a tier, its tier says so.
• Fails safe: unknown model → honest miss + acquisition demand, never a hallucinated procedure. Our own eval: zero hallucinated controls across the shipped-package benchmark.
• Re-expressed, never reproduced: packages are our own machine-structured re-expression; no manufacturer documents are redistributed.

Errors