spinabot/brigade

What it solves

Brigade provides a self-hosted, enterprise-grade personal intelligence system that allows users to manage a "crew" of AI agents. It solves the problem of fragmented AI interactions by centralizing memory, credentials, and tool access across multiple platforms (terminal, WhatsApp, Telegram, Slack, etc.) while maintaining strict data sovereignty and privacy.

How it works

Brigade operates as a gateway-client architecture. A long-lived gateway process manages the state, agents, and memory, while thin clients (like a TUI or messaging app adapters) connect via WebSockets. It uses a resilient agent loop to handle turns, routing messages to specific agents and sessions. For storage, it supports a local filesystem or a self-hosted Convex database. It integrates with various LLM providers (Claude, OpenAI, Gemini, Ollama) and can even leverage existing CLI subscriptions (like Claude Code) to avoid per-token API costs.

Who it’s for

It is designed for users who want a powerful, private, and extensible AI agent ecosystem that they host themselves, ranging from developers who prefer a terminal-first workflow to power users who want their AI agents accessible via common messaging apps.

Highlights

• Tideline Memory: A long-term memory engine with hybrid recall (keyword + vector), origin scoping to prevent data leaks, and automatic decay.
• Multi-Agent Org Chart: Ability to spawn isolated agents with their own personas and credentials, organized in a formal hierarchy that governs communication.
• Omnichannel Access: Integration with WhatsApp, Telegram, Slack, Discord, iMessage, and BlueBubbles.
• Extensive Tooling: Includes 1,000+ app connectors via Composio, the ability to read/write/edit Office documents and PDFs, and a deterministic motion-graphics video renderer (HyperFrames).
• MCP Support: Can expose its long-term memory as an MCP server or connect to other MCP servers.
• Privacy-First: No telemetry, no accounts, and local storage of API keys (mode 0600).