Cookwala

Cookwala / For

For device makers: ovens, hobs, multicookers, kitchen robots and humanoids

A device that implements Cookwala reads a signed recipe, checks every step against what it can do and sense, refuses what it cannot verify before anything heats up, and keeps its own safety limits whatever the recipe, the agent or the network says. Five endpoints, one capabilities document, public tests.

Why it matters to you

You already own the hard part: motion, heat control, sensing. Cookwala gives you the task layer above it, open and royalty-free: a recipe format a machine can plan, operation envelopes with physical bands, a sensor ladder per step, local safety limits that nothing can raise, and a refusal that is a normal answer, not an error. Your customers get every cuisine without you writing a recipe; your regulator gets a safety story they can read; your sales team gets a signed conformance report instead of a claim.

What exists today: the Core 0.2 API, a reference executor you can run beside yours, 51 Core conformance vectors including executor dry runs, a signed report format and a verifier. What does not: a certifier, a hardware test, a second implementation. You would be the first, and the standard would change where you show it is wrong.

Ways in, light to deep

  1. Light. Write a capabilities document for one product and dry-run the nine example recipes against it in the browser or the CLI; read Core 0.2 section 3 and 6 in an evening.
  2. Medium. Implement the five Core endpoints on a hub beside your firmware, run the executor suites, start the reference hub with a token and compare answers on the wire.
  3. Deep. Enforce the safety-limits pack on the device, measure the local stop, publish a signed conformance report, open a pull request that lists it in the registry, and tell us what to change in Core 0.3.

First success, under 15 minutes

Under 15 minutes, no hardware: run the executor suites, then start the reference executor with authentication on and dry-run a recipe against your own capabilities document.

git clone https://github.com/amado2k5/cookwala && cd cookwala && pip install -e sdk/python
python tools/run_conformance.py envelope   # 14 vectors: bands, altitude, sensor ladders
python tools/run_conformance.py dryrun     # 7 vectors: refusal before heat
cp examples/capabilities/robot-arm.json my-device.json   # edit ops, sensors, safety
cookwala dryrun examples/koshari.cookwala.json --device my-device.json --human-present
conformance: 14/14 passed · conformance: 7/7 passed · accepted or refused with the step and the reason

The path after that

  1. Write the capabilities document: operations, sensors with accuracy, safety (certifications, presence detection, stop)
  2. Dry-run the example recipes; every refusal names the step and the reason
  3. Implement POST /v1/executions, GET status, stop, resume and the log, plus GET /v1/capabilities and /v1/safety-limits
  4. Run the executor suites (envelope, dryrun, transitions) and the wire checks in the hub README
  5. Publish a signed conformance report; until a registry service exists, a pull request adds it to the registry file
  6. Nothing is certified yet: no certifier exists, and the report says what it proves and what it does not

Other paths