00brain / retrieval

v0.9.4 — generally available

Instant retrieval,
engineered for your data.

Connect a bucket, a website, or a database. Brain spins up a production RAG pipeline — embeddings, vector store, reranking, citations — and serves it from the edge in under 80ms.

Start building free Read the docs ↗

Trusted by

LinearNOTIONvercel ▲RetoolRAMP

brain ▸ acme-help

● live

How do I rotate API keys?

EMBED

bge-large-en

VECTOR SEARCH

cosine · top 8

RERANK

ms-marco

GENERATE

claude-haiku-4.5

01security/keys.md0.94

02guides/rotation.md0.88

03api/auth.ts0.82

ANSWER

▍

p50 retrieval

<80ms

global edge median

Indexes managed

12.4k+

across all customers

Tokens / month

4.1B

embedded & served

Recall@10

94.2%

on BEIR benchmark

— 01

Spin up RAG in minutes

Point Brain at your data source. We handle chunking, embeddings, vector storage, retrieval, and reranking. One integration, one endpoint.

— 02

Always fresh, never stale

Continuous re-indexing on a schedule, on webhook, or on file change. Your answers reflect what your data looks like right now — not last week.

— 03

Runs at the edge

Inference, embedding, and retrieval co-located on 320+ POPs. p50 under 80ms anywhere on the planet. No cold starts.

01Capabilities

Everything you need
to ship retrieval.

Six primitives that work together out of the box, but compose with any framework, any model, any vector store you'd rather bring yourself.

Continuously updated indexes

Tracks file changes, webhook triggers, and schedule-based re-indexing. Never serve stale data again.

Metadata filtering

Tenant-scoped, role-scoped, language-scoped. One Brain project, secure search contexts for every customer.

Web parsing source

Crawl your docs site or marketing pages. Auto-discovers sitemaps, respects robots.

Edge inference

Embeddings + LLM responses served from the closest POP. <80ms p50 globally.

NLWeb support

Open standard for natural-language web. Deep links to source content built in.

02RAG in action

One function call.
That's the SDK.

Search, filter, stream, and chain into agents — using the same primitive. Bring your own model or use ours. JS, TS, Python, Go, and HTTP.

search.ts

1import { initSearch } from #D4FF3C">"@aistack/search";
2
3const client = initSearch({
4  mode: #D4FF3C">"api",
5  api: {
6    baseUrl: #D4FF3C">"https://brain.aistack.run",
7    apiKey: process.env.API_KEY ?? #D4FF3C">"",
8  },
9});
10
11const result = await client.aiSearch({
12  query: #D4FF3C">"How do I rotate API keys?",
13  max_num_results: 8,
14});
15
16console.log(result.response);
17console.log(result.data);

● ready·78ms·8 chunks retrieved↗ run in playground

03Use cases

Built for the search problems worth solving.

SupportAnswer in your voice, with your sources

Customer-facing chatbots

Build a chatbot that answers customer questions using only your product content. Citations link back to docs. Refusals when confidence drops.

👤 How do I cancel my plan?

You can cancel any time from Settings → Billing. Your plan stays active until the end of the period.

↗ docs/billing/cancel · ↗ docs/refunds

p50 latency

94ms

Read the case study →

04Customers

We replaced three vendors, two cron jobs, and an entire team's worth of glue code with a single Brain endpoint. Search recall went up 38%, and we ship faster.

Lin Park

Head of AI · Lattice

05Platform

Powerful primitives,
seamlessly integrated.

Brain is the retrieval layer in the AIStack platform: a focused home for your documents, embeddings, search, and generated answers. Bring more pieces as you grow.