Overview

Relevant source files

LobeHub is an open-source, high-performance AI Agent framework designed to organize agents into a structured, 24/7 operating AI team package.json4-15 It moves beyond isolated, task-driven chat interfaces by treating Agents as the unit of work, providing infrastructure for multi-model intelligence, extensible tool use via the Model Context Protocol (MCP), and persistent personal memory README.md107-114

Core Feature Set

The framework is built around a comprehensive suite of AI capabilities:

Multi-Model Support: Unified access to various LLM providers (OpenAI, Anthropic, Google Gemini, DeepSeek, etc.) and modalities including vision and reasoning README.md135-140
Extensible Tooling: Integration with over 10,000 skills through a plugin system and full support for the Model Context Protocol (MCP) README.md138-142
Personal Memory: A "white-box" memory system that allows agents to learn from user workflows while maintaining transparency and user control README.zh-CN.md175-179
Multi-Agent Collaboration: Support for Agent Groups, enabling parallel collaboration, automated scheduling via agentCronJobs, and project-based organization README.zh-CN.md152-160 CHANGELOG.md181
Knowledge Base (RAG): Advanced document processing, chunking, and vector search for retrieval-augmented generation CHANGELOG.md63-142
TTS/STT: Built-in support for speech synthesis and recognition package.json14-15

Deployment Models

LobeHub is designed for flexible deployment across various environments:

Model	Description	Key Components
Cloud / Serverless	Optimized for Vercel, Zeabur, and Sealos using Next.js serverless functions.	`vercel.json`, `package.json:53`
Docker	Standalone containerized deployment for self-hosting with PostgreSQL and Redis.	`Dockerfile`, `docker-compose/`
Desktop	Cross-platform application (Windows, macOS, Linux) for local-first use and MCP stdio support.	`apps/desktop`, `package.json:61-68`
PWA	Progressive Web App support for mobile and desktop browsers.	`src/spa/`, `package.json:29-34`

Sources: package.json42-53 README.md60-63 CHANGELOG.md87-91

High-Level Subsystems Map

The following diagram illustrates the relationship between the major architectural components and their corresponding code entities.

System Context and Data Flow

Sources: package.json35-40 CHANGELOG.md150-155 README.md112-114

LobeHub is organized as a pnpm monorepo. To understand the system in depth, please refer to the following child pages:

Monorepo Package Structure

Details the layout of the packages/* and apps/* directories. It explains the responsibility of core logic packages (like agent-runtime) versus application entry points (like apps/desktop or apps/cli). For details, see Monorepo Package Structure.

Key Product Concepts and Mental Model

Explains how high-level product concepts such as Agents, Topics, Sessions, and Knowledge Bases map to specific database schemas (managed via Drizzle) and Zustand store slices. For details, see Key Product Concepts and Mental Model.

System Relationships: Natural Language to Code Entity

Sources: package.json35-40 CHANGELOG.md63-153