Plans

Metered on what it costs to answer.

Two units, and both of them are things you can predict. An agent turn is one model round on /ask. An API request is one call to the REST or MCP surface. Tool calls and ledger reads are recorded and never billed.

Free

£0

Evaluate it properly. Every tool, both chains, and the same verification stamps a paying customer gets.

Rate: 10/min
Agent turns: 50 / day
500 / month
API requests: 1,000 / day
10,000 / month

Start free

Developer

most teams

Pricing in progress

Room to build against. Enough headroom for an agent in a loop, a backfill, or a reconciliation that runs nightly.

Rate: 60/min
Agent turns: 500 / day
8,000 / month
API requests: 25,000 / day
500,000 / month

Talk to us

Enterprise

Let's talk

Custody and exchange scale. Negotiated limits, the chains you actually settle on, and a reconciliation we shape with you.

Rate: Negotiated
Agent turns: Negotiated
API requests: Negotiated

On every plan

Provenance on every balance change
Verification standing on every answer
Bitcoin and Ethereum, mainnet
The full tool surface — no tools held back
MCP and the /ask console

What we don't charge for, and why.

Not per tool call

Six fast coverage calls would cost more than one thirty-second aggregate — backwards on spend, latency and load at once. It would also push the model toward being less thorough on your question, which is the opposite of what you are paying for.

Not per ledger read

Answers at or below the balanced watermark are immutable, so we cache them. Billing per read would make your invoice depend on a cache you cannot see or reason about.

Not per verification

Reconciliation against the node's own commitment is the product, not an upsell. Every plan gets the same stamps — clean-verified, clean-unverified, tainted with scope — because an answer without its standing is not a cheaper answer, it is a worse one.

How the limits behave.

Periods are UTC days and months: Not rolling windows and not your billing anniversary. A day resets at 00:00 UTC, a month on the 1st.
Rate limits are per key: One runaway key cannot spend the whole account's per-minute budget. Sessions and the anonymous demo get their own bucket.
429s tell you when to come back: Retry-After is the real number of seconds until the window that stopped you rolls over — not a fixed guess.
A rejected request is not billed: Reserving before checking is how a concurrent burst is counted honestly; anything refused is credited straight back.
Signed out still works: The console answers 20 turns a day without an account, on the same tools and the same verification. Signing in raises it.

Start on the free tier.

Start free Talk to us