Your AI has amnesia.

Every architecture decision, every debug session, every preference you expressed — gone between sessions. BrainLayer gives any MCP agent persistent memory that survives restarts, compactions, and context limits.

Get started View source

Free · open source · local-first · one SQLite file

Without BrainLayer

×Repeats the same mistakes every session
×Forgets architecture decisions overnight
×No context survives a restart or compaction
×Rediscovers bugs it already fixed last week
×Asks you the same clarifying questions again

With BrainLayer

&check;Remembers every decision, learning, and correction
&check;Searches 294K+ knowledge chunks in under 50ms
&check;Knowledge graph connects entities across sessions
&check;Cross-session memory persists through restarts
&check;Recalls your preferences, patterns, and past work

The Problem

Your AI forgets everything between sessions

Every architecture decision, every debugging session, every correction you gave it — gone. BrainLayer gives any MCP agent persistent memory backed by semantic search and a knowledge graph.

Semantic search

Hybrid vector + keyword search with RRF

brain_search combines bge-large embeddings with FTS5 keyword matching via Reciprocal Rank Fusion. One query searches across every conversation you have ever had. Sub-50ms on 300K+ chunks.

Knowledge graph

Entities and relations that grow over time

brain_digest extracts entities, relations, and action items from raw content. brain_entity looks up any entity in the graph with evidence and connections.

Persistent memory

Decisions, learnings, corrections — stored forever

brain_store persists any memory with auto-type detection, auto-importance scoring, and per-agent scoping. Chunk lifecycle management (supersede, archive) keeps knowledge current without losing history.

Companion App — macOS

BrainBar

Optional 209KB native Swift menu bar app. Quick capture, live dashboard, knowledge graph viewer — all over a Unix socket.

Requires the BrainLayer MCP server.

MCP tools

Twelve working tools. One memory layer.

Core - what you use daily

brain_searchHybrid semantic + keyword + KG search with compact formatted output

brain_storePersist decisions, learnings, corrections, and capture notes

brain_recallUnified recall entrypoint for search, entity, and session-aware lookup

Advanced

brain_entityKnowledge graph entity lookup

brain_expandDrill into one search hit with surrounding context

brain_digestDeep-ingest large content and extract entities, actions, and relations

brain_updateUpdate chunk importance and tags by chunk ID

brain_tagsList unique tags with counts, filter by prefix

brain_get_personLook up a person entity with all known relations

Lifecycle

brain_supersedeReplace a chunk with a newer version, preserving history

brain_archiveSoft-delete a chunk with timestamp for audit trail

brain_enrichTrigger deep enrichment on a stored chunk via Groq or Gemini

Works with

Any MCP client

Claude Code

Cursor

Zed

VS Code

Codex

Kiro

Gemini CLI

Getting started

Three steps

Install from PyPI

Configure MCP and create the local database

Start indexing and optional BrainBar capture flows

BrainLayer runs local-first. MCP clients talk to the same memory layer, and BrainBar exposes the same formatter and database through a menu-bar capture flow on macOS.