User Guide

Everything AIBrain can do — memory, workflows, MCP, mesh networking, evolution, and more.

Core Concepts

Memory System

Storage Architecture

• SQLite + FTS5 — Full-text search across all memories, instant results
• Vector embeddings (optional) — Semantic search using MiniLM (22MB) or bge-base (110MB)
• Hyperedge relationships — Multi-node connections, not just pairwise links
• Entity extraction — Auto-detects technologies, URLs, emails, IPs, names

Memory Types

Search Modes

AIBrain uses selective query routing — it auto-detects your query type and routes to the best search strategy:

• Keyword queries → FTS5 full-text search (fast, exact matching)
• Semantic queries → Vector similarity search (meaning-based)
• Hybrid queries → Both, with result fusion

$ aibrain search "kubernetes deployment"     # Search memories
$ aibrain summary                            # Memory stats
$ aibrain brain export                       # Export as JSON
$ aibrain brain import backup.json           # Import from file
$ aibrain forget "outdated topic" --confirm  # Delete memories

CLI Reference

Setup & System

Memory & Brain

Stack & Versioning

Workflows

AIBrain ships with 55+ workflow scripts organized into categories: Productivity, Research, Communication, Developer, Self-Learning, and System. See the Workflow Reference for the full list.

$ aibrain workflows list                  # Show all workflows + status
$ aibrain workflows enable --recommended  # Enable the starter set
$ aibrain workflows sync                  # Create OS scheduled tasks
$ aibrain workflows run weekly_wins       # Run one immediately

Workflows use your OS scheduler (cron on Linux, schtasks on Windows, launchctl on macOS). No daemon process required.

Brain Packs

$ aibrain packs                       # Browse available packs
$ aibrain packs activate developer     # Activate a pack
$ aibrain packs active                 # Show active packs

Dashboard

Launch with aibrain server. The dashboard runs at http://localhost:8001 and includes 17 pages:

Additional pages: Activity, Settings, Logs, Skills, Backups, Webhooks, MCP Hub, and Integrations.

Keyboard shortcuts: Ctrl+K opens the command palette. ? shows all shortcuts.

MCP Servers

AIBrain ships with 9 MCP servers, making it one of the most comprehensive agent toolkits available. Every server is agent-agnostic — works with Claude, GPT, Cursor, Copilot, Ollama, or any MCP-compatible runtime.

# Install all MCP servers at once
$ aibrain mcp install

# Or add individually
$ claude mcp add aibrain-memory -- python mcp_server.py --db aibrain.db
$ claude mcp add it-support -- python mcp_it_support.py
$ claude mcp add peer-discovery -- python mcp_peer_discovery.py

Multi-Agent Mesh

Multiple AIBrain instances can coordinate through the agent mesh. Agents register with a broker, exchange messages and files, sync memories, and resolve conflicts through consensus merge.

$ aibrain mesh status                        # Show mesh agent status
$ aibrain mesh register myagent agent.db      # Register an agent
$ aibrain mesh merge                          # Merge all agents (consensus)
$ aibrain mesh distribute                     # Push merged brain to all

Swarm Training

Spin up multiple agent instances, let them learn independently, and harvest the results:

$ aibrain mesh swarm create research -n 100   # Create 100-agent swarm
$ aibrain mesh swarm harvest research          # Merge swarm results

Self-Learning & Evolution

AIBrain agents improve over time through built-in self-learning workflows:

• Correction Learner — converts your feedback into permanent skill principles
• Memory Hygiene — deduplicates, decays stale data, keeps the brain fast
• Self-Model Refresh — rebuilds the agent's understanding of its own strengths weekly
• Principle Validator — tests stored principles against actual outcomes
• Contradiction Detector — finds and resolves conflicting memories
• Evolution Experiment — runs self-improvement experiments on a schedule

These run silently in the background. You don't interact with them directly — they just make everything better over time.

Marketplace

Package and sell trained brain configurations. Buyers get domain expertise instantly without the training time.

$ aibrain marketplace package --name "devops-brain"  # Package for sale
$ aibrain marketplace publish devops-brain.brain     # Publish
$ aibrain marketplace search "machine learning"      # Browse
$ aibrain marketplace install <id>                   # Install a brain
$ aibrain marketplace revenue                        # Revenue dashboard

Tiers & Pricing

$ aibrain license tiers              # See pricing details
$ aibrain license activate <key>     # Activate your license