Credit model
Why search costs 0.5, extract costs 1 per URL, research costs 5 — and what that actually buys.
Credit pricing on Brime is per call, fixed. No per-token billing and no surprise charges. This page explains the reasoning so agents and finance teams can reason about cost without screen-scraping the dashboard.
The table
| Endpoint | Credits | What it covers |
|---|---|---|
POST /v1/search | 0.5 | One web search call with ranked results and an optional synthesised answer. |
POST /v1/extract | 1 per URL | One clean-markdown extraction per successfully extracted URL; failed URLs are free. |
POST /v1/research | 5 | The full multi-round research agent, with citations. The Tavily-compatible deep research path costs 15. |
GET /v1/research/{job_id} | 0 | Polling never bills. |
GET /v1/research/{job_id}/stream | 0 | Streaming and resume via Last-Event-ID never re-bill. |
GET /v1/usage | 0 | Account introspection. |
Free tier: 1,000 credits per month. Paid plans buy larger monthly buckets at falling marginal cost. See Pricing.
Why these numbers
The unit cost is shaped so the call you'd reach for is also the cheapest one that satisfies the intent:
- Search (0.5) — the everyday call. Ranked results plus an optional synthesised answer. Use it freely.
- Extract (1/URL) — fetch and clean a page. We charge a flat 1 per successfully extracted URL regardless of whether a static fetch was enough or a full browser render was required. Failed URLs are free.
- Research (5) — the full multi-round agent. Internally it does the work of several search and extract calls plus a synthesis pass; 5 credits is meaningfully cheaper than wiring that loop yourself.
Errors do not consume credits. A 401 / 429 / 502 response carries
credits_used: 0. Partial-failure refunds are automatic and surface
on the same response.
What credits_used on the response means
Every response includes credits_used. It is the debit applied for
this single call — final by the time the response leaves Brime.
There is no async settlement: your balance is decremented before the
response reaches you, so the value you see is authoritative.
Real-time balance
Each response carries X-Credits-Used and X-Credits-Remaining headers
so a client can budget in flight. The dashboard at
brime.dev shows the same balance.
What does not affect cost
- Response size. Search is 0.5 whether you ask for 1 result or 20.
- LLM token usage. Synthesis is included in the flat call price.
- Browser escalation in extract. Tough pages that need a real browser cost the same 1 credit per URL as easy static ones.
- Retries from the SDK. Brime's SDKs retry transient 5xx automatically before surfacing a response — only the final successful (or terminally errored) response is charged.
- Idempotency replays. A cached idempotent response is served without debiting credits again.
See also
- Pricing — plan tiers and monthly buckets.
- Rate limits — what your QPS cap looks like once you have credits to spend.
- Idempotency — why a retried call
with the same
Idempotency-Keydoesn't double-bill.