Free key

A key in one step, no card

Teach a model something, ask it, delete it again. The free tier is meant to be enough to wire into something real, not a tour.

Checking whether the hosted API is live…

What free includes

LimitFreeWhy this number
Questions10,000 a month Naming the rule makes no model call at all, so these are nearly free to serve.
Teaching a rule500 a month A write is a fifth of a second of gradient descent, and it holds the model while it runs.
Memories50 Kept well inside the range where finding a memory from plain English still works. See below.
Rate5 a second The limit that actually protects the service. A month's worth spent in a minute is an outage for everybody else.

Past the ceiling you get 402 with the limit you hit and when it resets — not a throttle, because a throttle looks like an outage. Deleting is never refused for quota, at any tier, for any reason.

Sixty seconds

export KEY=hebb_...        # from above

# teach it something
curl -X POST $BASE/v1/rules -H "Authorization: Bearer $KEY" \
  -H 'content-type: application/json' \
  -d '{"slot": "refund window", "value": "60 days"}'

# ask, naming the rule
curl -X POST $BASE/v1/ask -H "Authorization: Bearer $KEY" \
  -H 'content-type: application/json' \
  -d '{"question": "how long to refund?", "slot": "refund window"}'
# {"value": "60 days", "confidence": 1.0, ...}

# forget it, and get a receipt proving it
curl -X DELETE "$BASE/v1/rules/refund%%20window" -H "Authorization: Bearer $KEY"

# what you have spent
curl $BASE/v1/usage -H "Authorization: Bearer $KEY"

Pass slot when you know it

If your application already knows which rule it wants, name it. That path returns the stored value exactly and makes no model call. Leaving it out asks the memory to work out which rule you meant from the question, which is approximate today — it is the part we are still improving, and we would rather say so than have you find out.

The limits are not the pitch

The free tier runs on our hardware, so it is metered. The product most people buy is the self-hosted licence: the same container inside your own perimeter, your data never leaving it, and no per-request ceiling because it is your GPU. Ask about that once the free key has convinced you.