Brainby arc-labs/docs
Brain on GitHubDashboard →
Guides
Guide

Guides

Task-shaped walkthroughs — adding memories, recall, building context, streaming, idempotency, multi-tenant, error handling, and more.

Core paths

write

Adding memories

Encode text with encode (POST /v1/memories), read the structured outcome, and control content dedup and event time.

read

Recall

The one read verb: a membership verdict — Single, Many, or None — over the caller's memory pool.

context

Building context

Assemble recall output into a prompt block yourself — Brain has no build-context verb.

streaming

Streaming

The wire client's *Frames methods drain a response frame-by-frame; HTTP has none.

Production patterns

→ resilience
Idempotency

request_id idempotency and content dedup (BLAKE3 fingerprint, was_deduplicated) — two mechanisms, two jobs.

→ resilience
Error handling

BrainHttpError vs the BrainError hierarchy, the nine categories, and which are transient.

→ resilience
Retries & cancellation

The HTTP retry policy, safe wire retries via the request-id contract, and timeouts.

→ auth
Identity walkthrough

How a key binds a (namespace, agent) scope, and how to confirm it with whoami.

→ auth
Multi-tenant patterns

Namespace-per-tenant isolation, one key per agent, and a pooled key with act_as.

→ ops
Namespace lifecycle

Plan, create, populate with schema, and decommission a tenant boundary.

→ advanced
Graph queries

When to reach past recall into the typed entity/relation graph.

Was this page helpful?

Architecture

Brain is one Rust binary with a strict one-way crate graph — an mmap arena, a WAL, redb, HNSW, tantivy, and bundled models, sharded thread-per-core. No external database.

Adding memories

Store text with encode (POST /v1/memories), read the structured outcome, and control content dedup, event time, and idempotent retries.

On this page

Core pathsProduction patterns
Suggest an editOpen in Playground