Pricing and plans

The tables below are generated from the tier definition and the price list the engine enforces, so the credits on this page are the credits your key is charged. Manage your plan in the console.

Plans

The free tier is deliberately generous — 10,000 credits a month with no card, which is 10,000 searches or 2,000 ground calls. The per-minute rate is what does the work of separating prototyping from production: 60 credits a minute is enough to build and evaluate against, and not enough to serve real users from. A workload that outgrows a prototype hits the ceiling on latency rather than on the allowance, which is the more honest signal.

What a call costs

Every call costs a fixed number of credits, known before it runs and the same whatever it returns. A limit=0 count costs what limit=100 does, and ground costs 5 however many stories it expands on the server.

• A call that runs and fails costs 1, whatever it would have cost: a 400, a 404, a timeout, a 503.
• A refused call costs nothing: 401, 402, and both kinds of 429.
• Every response says what it was billed in x-credits-charged. /describe lists this table, and every operation in the API reference carries its cost as x-credits.

The per-minute rate is counted in credits too: a call takes its cost from the minute. On Free that is 60 searches a minute, or 12 ground calls.

Over MCP a tool call costs what its REST operation costs. Connecting is free: initialize, the initialized notification, tools/list and ping take no credits and no per-minute tokens, so opening a session costs nothing.

These are free and need no key: /describe, /openapi.json, /mcp/tools.json, /llms.txt, /guides/…, and the health endpoints.

/account costs no credits but does need a key — what it reports is that key's plan, allowance and usage, so there is nothing to answer without one. It still takes one per-minute token, like any call, and it answers even when a hard-capped key has run out.

When you run out of credits

On Free the key stops: a call that costs more than the credits left answers 429 until the period rolls over, with no retry-after, because retrying will not help. The refusal costs nothing, and a cheaper call that still fits keeps working. There is no card on file, so free means free.

On the paid plans calls keep working past the allowance. The pricing page promises that rather than failing closed — an agent in production should not stop because a counter rolled over. Overage pricing is not yet published; if you expect to run past your allowance regularly, talk to us before you do rather than after.

History depth

How far back a plan may query. A tier with a 30 days window sees the recent index; Archive sees all of it. This is a filter on what a query can reach, not a filter on what exists.

Spending less

• limit=0&facets=true sizes a topic in one search instead of five exploratory ones.
• collapse=story stops you buying — and reading — the same article five times.
• fields trims the response to what you will actually use.
• assemble_context is one 3-credit call in place of a retrieval loop that would be several.
• The snippet is the passage text, so get_document is rarely needed.
• A failed call still costs 1, so validate parameters before a batch rather than discovering a typo a thousand times.

GET /account reports your plan, per-minute rate and credits used this period. See Rate limits and credits for what the refusals mean.

How these rates compare with other search APIs, and the unit economics behind them, are on the main site: pricing, the 2026 landscape and the cost model. Every competitor figure there carries a dated link to the vendor's own pricing page.