An open-source multi-agent collaborative AI assistant built in Go, featuring 8 specialized agents, 4 working modes, MCP protocol support, and multi-platform deployment.
fkteams (FeiKong Teams) is a robust multi-agent collaborative framework designed to solve complex tasks through specialized division of labor. The project features 8 built-in dedicated agents for coding (@XiaoMa), command execution (@XiaoLing), searching (@XiaoSou), data analysis (@XiaoXi), remote management (@XiaoFang), creative writing (@XiaoShuo), summarization (@XiaoJian), and general-purpose orchestration (@XiaoZhu), along with four operating modes: team collaboration, deep search, roundtable discussion, and custom mode. Built on the Cloudwego Eino framework, it natively integrates the Model Context Protocol (MCP) for flexible external tool invocation.
For interaction, fkteams provides both a modern Web interface and a Bubble Tea-based terminal UI (TUI). It supports multimodal input (text, images, audio, video, files) and streaming display of reasoning processes from models like DeepSeek-R1 and o1/o3. The system features cross-session long-term memory and implements highly reliable execution mechanisms including sub-task parallelism, SSE-based streaming task control with reconnection, and automatic output truncation recovery with JSON repair. Models can proactively ask users interactive questions (single/multi-select + free input), and scheduled tasks can be set via natural language.
For external integration, fkteams exposes an OpenAI-compatible API, supports one-click OAuth login to GitHub Copilot, and can connect to IM platforms like QQ and Discord. Users manage multiple LLM providers through a TOML configuration file, create custom agents via [[custom.agents]], and dynamically enhance capabilities through the Skills system. The project supports one-click script installation, Docker containerized deployment, and source compilation, balancing ease of use with deployment flexibility. Latest version: v0.20.3, MIT licensed.