AIBrain is the AI agent memory layer that compounds across every session. It gives any agent a persistent memory layer — plus Agent/Task/Team composition, a Flow engine, document ingestion, universal MCP memory server connectivity, 71 self-learning workflows, mesh networking, and a 60-page native dashboard. Install a brain pack and your agent has years of expertise in minutes.
aibrain setupPick what fits your life. Each pack is a complete brain transplant — workflows, skills, knowledge, and personality.
Everything you need to start
Single instance • 71 workflows • 9 MCP servers • 60-page dashboard • Local SQLite • No credit card
Install FreeFor power users and professionals
Agent Teams • Flows • Document Ingestion • MCP Client • Boss Agent • Model routing • Skill evolution
Get ProMulti-machine, multi-agent
Up to 10 workers • Cross-machine via Tailscale • Team brain sharing • Priority support
Get TeamOther tools give you an empty notebook. AIBrain is a complete AI agent memory operating system — a local MCP memory server, 9 built-in MCP servers, workflows, mesh networking, and a dashboard. Whether you need LangChain persistent memory, an agent memory layer for CrewAI, or a universal brain for any MCP client — it ships out of the box.
Brain API — add memories in 3 lines of code. Automatic fact extraction from conversations, memory deduplication (no duplicates ever), per-user memory scoping, and full change history tracking. Powered by SelRoute, the research-backed retrieval engine validated on 62,000+ instances.
Agent Teams with task routing, org charts, company structure, and goal tracking built in.
Decorator-driven state machines. @start, @listen, @router decorators build execution graphs with conditional routing. Typed state, persistence for long-running flows, human-in-the-loop pause/resume, @retry, @timeout, and visualization.
brain.ingest("file.pdf") — feed documents directly into memory. 7 source types: PDF, CSV, Excel, JSON, JSONL, Text, Markdown, and directories. Sentence-aware chunking. Incremental — re-running skips already-processed files.
Connect to any MCP server from Python. Three transports: stdio, HTTP, SSE. Tool bridging turns remote MCP tools into local callable functions. Registry manages multiple servers from code or the dashboard.
Built-in web search — no API keys needed. Your agent can search the web, fetch pages, and summarise results. Zero config, works immediately.
Install a pack and your agent has years of expertise in minutes. Workflows, skills, knowledge, and personality — 8 domains pre-configured.
Built-in system diagnostics, service monitoring, and automated remediation. 9 tools + 8 workflows + ticketing system. Your agent is also your IT department.
Encrypted agent-to-agent communication over Tailscale. Send files, sync brains, share skills between devices — no port forwarding, works across NAT.
Install AIBrain on your phone as a Progressive Web App. Full dashboard, push notifications for approvals, offline access. No app store needed.
Built-in AI code editor with live preview. Write, debug, and deploy code with your agent's help.
Manage calendars (CalDAV + local), search and organize files, find duplicates, auto-classify documents. Your agent handles the busywork.
Generate social media images, banners, and thumbnails programmatically. Built-in templates for LinkedIn, X, YouTube. Your agent creates your content visuals.
Real-time LLM cost tracking by model and task. Know exactly what each workflow costs.
Your agent scores every run, extracts learnings, and improves automatically. Compounds across sessions.
Automated optimization loops — hypothesise, modify, evaluate, keep or revert. Your agent improves its own performance using the scientific method.
Everything runs on your machine. SQLite database, no cloud dependency. Works with any LLM — Claude, GPT, Gemini, Llama, Ollama. No vendor lock-in.
From zero to a trained brain in three steps.
One command. The setup wizard handles Python, dependencies, database, config, and MCP registration automatically.
Choose your domain. Workflows, skills, knowledge, and agent personality all pre-loaded and ready to run.
aibrain start — stack restores, services launch, dashboard opens. Your agent is ready.
Everything your agent needs to be genuinely useful. No assembly required.
The core agent memory layer — add, search, and manage memories in 3 lines of code. SQLite + FTS5 full-text search with optional vector embeddings. Works as a drop-in LangChain persistent memory backend or standalone Brain API. Powered by SelRoute, a research-backed hybrid retrieval engine published on arXiv.
Typed multi-agent composition — define agents, create tasks, run teams. Sequential or hierarchical processes with delegation and guardrails.
Decorator-driven state machines for complex, multi-step workflows. Build execution graphs with conditional routing, persistence, and human-in-the-loop support.
Feed documents directly into brain memory. brain.ingest() handles 7 source types with intelligent chunking and incremental processing.
Connect to any MCP server from Python. Three transports, tool bridging, and a registry for managing multiple servers.
Pre-built automations across six registries. Enable any workflow with a single toggle. All schedules are configurable via cron expressions.
Full MCP server suite that any compatible agent can connect to. Each server is a focused capability module.
Encrypted peer-to-peer communication between AIBrain agents on any device, anywhere. No port forwarding. Works across NAT.
Your agent stays connected. Telegram bot notifications for approvals and alerts. IMAP/SMTP email monitoring with auto-reply capability.
Full browser UI for everything your agent does. Not a toy dashboard — 45+ pages with teams, flows, ingestion, MCP management, and real-time WebSocket updates.
Your agent improves itself. Outcomes feed patterns, patterns generate hypotheses, hypotheses run experiments, experiments get measured and kept or reverted.
One command gets you from zero to running. The wizard detects your environment and configures everything automatically.
Build a team of AI agents that work together, learn together, and get better every day.
CEO delegates to managers, managers assign to workers. Real org chart with reporting chains, roles, and responsibilities — just like a human company.
Create tasks, assign to agents, track progress. Agents claim work, execute via LLM, submit for QA review. Quality gating ensures nothing ships without meeting standards.
Set company goals with milestones and deadlines. Track progress across the whole team. Sub-goals break big objectives into agent-sized work.
Agents wake on schedule, check for work, execute tasks, and learn from the results. CEO agents can delegate and create new tasks for idle workers.
Store API keys and credentials securely. Encrypted at rest, scoped per agent or company-wide. Agents access only what they need.
Every agent feeds the same brain. Knowledge from every task compounds. The whole company gets smarter with every heartbeat.
Claude, GPT, Gemini, Ollama, Groq, Together, DeepSeek, or your own local server. Each agent can use a different provider.
Simple tasks use cheap models ($0 with Ollama). Complex tasks use Opus. Automatic routing minimizes cost.
Quality reviews, hire approvals, escalations. Nothing happens without the right authority level.
60-page dashboard with teams, flows, ingestion, org charts, agent detail, task tracking, cost analytics, and evolution monitoring.
Evolution engine analyses outcomes, proposes experiments, and applies what works. Your company gets better autonomously.
Owner, admin, member, viewer. Invite team members with the right permissions. API key auth for external access.
AIBrain sits between your agent and everything it needs. One brain layer, any agent.
AIBrain is agent-agnostic. It works as an MCP memory server for any MCP-compatible client, and as a native persistent memory layer for LangChain, LangGraph, LlamaIndex, OpenAI Agents SDK, and CrewAI.
Your data never leaves your machine. Not even a little bit.
SQLite database on your filesystem. No cloud. No syncing to someone else's server.
Zero analytics, zero tracking, zero phone-home. We don't know you installed it.
Bring your own API keys or use Ollama for fully offline operation. No vendor lock-in.
Copy one file to move your entire brain. Export as JSON, import anywhere. You own it.
Everything you need to get started, build workflows, and extend AIBrain.
Install AIBrain, run the setup wizard, and launch your first brain pack in under 5 minutes.
Complete guidebook covering memory, workflows, brain packs, dashboard, mesh networking, and more.
All 71 workflows documented. Triggers, schedules, configuration options, and customization.
165+ REST endpoints. Memory, workflows, approvals, chat, agents, health checks, and more.
Free tier is genuinely useful. Pro unlocks the full brain. Team scales it.
| Capability | AIBrain | Typical alternatives |
|---|---|---|
| Persistent memory | Full API + vector + graph + satellites | Partial or cloud-only |
| Agent/Task/Team composition | Typed agents, sequential + hierarchical, delegation, guardrails | Manual setup or limited |
| Flow engine | Decorator-driven state machines with routing, persistence, HITL | Build your own or use external tools |
| Document ingestion | 7 formats, sentence-aware chunking, incremental | Separate tools required |
| Universal MCP client | 3 transports, tool bridging, registry | Single protocol or none |
| Self-learning workflows | 71 included | Build your own |
| Self-improvement / evolution | Full engine | Not available |
| Multi-agent companies | CEO, managers, workers, 8 LLM adapters | Basic multi-agent or cloud-only |
| Dashboard / UI | 45+ pages, fully local | Cloud UI or none |
| MCP servers | 9 servers (15 built-in tools) | 0-1 servers |
| Knowledge graph | Interactive force-directed | Rare |
| 100% local / private | Yes — your machine, your data | Often cloud-dependent |
| Research-backed retrieval | SelRoute (arXiv-published) | Standard retrieval |
One line. Copy it. Run it.
aibrain setup