Glossary

Brime-specific

Credit — Brime's billing unit. Each endpoint has a flat per-call multiplier: 0.5 for search, 1 per successfully extracted URL for extract, 5 for research. See Pricing for the full table.

Idempotency-Key — A header you (or the SDK) attaches to extract and deep research calls so retries don't double-charge. The Python and Node.js SDKs auto-generate one per call.

Drop-in — Brime's adapters expose Tavily, Exa, and Parallel wire formats under separate paths (/tavily/*, /exa/*, /parallel/*). Swap your client's base_url and your existing code keeps working.

Tier 0 / Tier 2 — Extract dispatch tiers. Tier 0 is the Worker-side fast path (HTML/PDF/DOCX). Tier 2 escalates to a full Chrome render when bot challenges or SPA hydration require it.

Standard terms (in Brime context)

SERP — Search Engine Results Page. Brime's gateway calls a SERP provider, then ranks and synthesises on top of the raw results.

RAG — Retrieval-Augmented Generation. Brime is built to be the retrieval half of a RAG pipeline: search → rank → feed an LLM.

MCP — Model Context Protocol. Anthropic's open spec for plugging tools into LLM clients. Brime ships an MCP server (mcp-server-brime) so Claude Code, Cursor, etc., can use Brime out of the box.

SSE — Server-Sent Events. Brime's research endpoint streams agent events (tool calls, sources, completion) over SSE; supports Last-Event-ID resume.

Last-Event-ID — A standard SSE header that lets a client reconnect to a stream and replay events from a known point. Brime's research stream supports this.

Idempotency — Running the same operation twice produces the same result. Brime's extract endpoint requires an Idempotency-Key so retries are safe.

Research — Multi-round iterative agent: plan queries → search → read sources → synthesise → repeat. Brime runs a single tuned flow, capped at 5 rounds and ~90 s wall-clock.