DEPRECATED — PROJECT ARCHIVEDthe-brain is no longer maintained. Source and documentation are preserved for reference only.
🧠the-brain

the-brain

Deprecated project archive — 3-layer cognitive architecture reference

the-brain archive

the-brain is a deprecated project archive. This documentation preserves the former 3-layer cognitive architecture and extension-first design for reference.

DEPRECATED — PROJECT ARCHIVED. the-brain is no longer maintained. Do not install or use it for new projects. Source code and documentation are preserved for reference only.


What the-brain Was

AI tools forget between sessions. the-brain was designed to remember across the tools you used. It harvested conversations, filtered noise from signal, and consolidated what mattered. No cloud. No telemetry. Just a background daemon learning your patterns.

InstallationHistorical installation notes; no supported installer
QuickstartHistorical runtime walkthrough
⚙️ ConfigurationConfig file reference
🏗️ Architecture3-layer cognitive architecture
🧩 Writing PluginsHistorical plugin extension guide
💻 CLI ReferenceComplete command reference
🌐 MCP ServerUse with Claude Desktop, Cursor, Zed
📡 Remote ModeHistorical Linux server and macOS client reference

Architecture at a Glance

The former architecture fed AI tools into three plugin slots:

Any AI tool → ⚡ INSTANT slot — immediate context (default: graph-memory)
            → ⚖️ SELECTION slot — retained information (default: spm-curator)
            → 🌌 DEEP slot — long-term consolidation (default: mlx-lora + auto-wiki)

Each slot was replaceable. The daemon loaded configured extension names from config.json ("extensions": ["my-ext"]); extensions were disabled by default for security.

What the-brain Was Designed to Learn

  • Corrections — "No, actually use useCallback here" → remembered for later
  • Preferences — "I prefer arrow functions" → injected into later contexts
  • Patterns — Repeated choices across sessions → suggested related concepts
  • Identity — Stable preferences and style tracked across retrains

Documentation Sections

SectionDescription
Start HereArchived installation, quickstart, configuration
Core ConceptsArchitecture, hook system, memory layers
CustomizationPlugins, harvesters, training, extensions
ReferenceCLI, config schema, MCP tools, env vars
IntegrationsMCP server, remote mode, AI tools, team mode, menu bar
DevelopmentArchived contribution guidance, project structure, testing

Tech Stack

ComponentTechnology
RuntimeBun + TypeScript
Databasebun:sqlite + Drizzle ORM
Pluginshookable event system
TrainingApple MLX (Python sidecar)
DocsNext.js + Fumadocs

License

MIT © 2026 Oskar Schachta

On this page