SudoJacky/tinybot

Tinybot is a local-first framework designed to serve as a personal AI assistant that connects large language models (LLMs) with a range of chat platforms, tool systems, and automation mechanisms. The framework is structured around a "service-first" architecture, where persistent state and complex orchestration are managed by Python services, while interfaces such as the command-line interface (CLI), web user interface (WebUI), and application programming interface (API) act as control planes. Tinybot is intended for users seeking to develop or operate AI-driven agents that can evolve, collaborate, and perform autonomous tasks across multiple environments.

Among its notable capabilities is an interactive agent loop, which assembles prompts, executes tools, and maintains session persistence through dedicated classes like AgentLoop and AgentRunner. Tinybot records the outcomes and errors of tool executions in an ExperienceStore, enabling the system to learn and improve performance over time. The Cowork multi-agent framework supports autonomous team architectures—including swarm, team, and generator-verifier models—managed by the CoworkService, facilitating collaborative and distributed agent workflows.

For knowledge management, Tinybot features a GraphRAG knowledge base, which offers a multi-stage pipeline for document ingestion, extraction of entities and relationships, and hybrid retrieval through the KnowledgeStore. Background automation is supported by built-in services for scheduled tasks and periodic check-ins, such as CronService and HeartbeatService. The system also provides a multi-channel gateway, integrating with WebSockets for its WebUI as well as third-party chat platforms including Feishu (Lark), DingTalk, and WeChat.

Tinybot is organized into distinct functional layers that separate presentation from core logic, with dedicated modules for agent execution, collaboration, knowledge management, configuration, and security. Users can interact with the system via CLI, API, or the WebUI, and the documentation references installation via Python tools and onboarding through a guided wizard. The framework is positioned as a lightweight, extensible platform for building agentic environments capable of self-evolution, multi-agent collaboration, and advanced knowledge retrieval.