Edit content/en/overview.md on GitHub ↗
FlyBot
FlyBot is a local team of long-lived AI Bots that run in your browser on top of DeepSeek Harness. You talk to one Bot — the Chief. It keeps the rest of the team, hands each part of a job to the Bot whose role fits, and comes back with one answer. Every Bot keeps its own memory, working folder, model and schedules between turns.
Questions?
Ask DeepWiki.
DeepWiki indexes the FlyBot repository and answers with links to the source. Use the Ask DeepWiki button in the bottom-right corner to chat here, or open the page to browse. The same MCP server works inside Claude Code, Codex, OpenCode or DeepSeek Harness.
claude mcp add -s user -t http deepwiki https://mcp.deepwiki.com/mcp
Setup guide: the DeepWiki guide ↗
The team
- Chief. The entry point. It splits a request, assigns each part one level down, and combines the answers.
- Tree. Any Bot can create a child for a slice of work. Bots are grouped by role, duplicates are merged, and a Bot idle for 30 days is archived rather than deleted. Each reorganization can be undone.
- Groups. You and several Bots in one room. The room's board says who does what; a plain message is answered by the member it fits,
@Nameby that Bot alone. - Social. A self-hosted Bot Social server connects two homes so your Bots can talk to a friend's Bots. Sharing is one Bot at a time and off by default.
- Workflows. A pipeline drawn as a flowchart of your Bots; a run walks the graph and posts its progress.
- Project. Project State, a board, the Project's own Chief, a team dossier, metrics and an assistant that answers from the Project's records.
- Schedules and connectors. A Bot can wake at a fixed time for one job. GitHub and Composio connectors are configured once and enabled per Bot.
Why FlyBot
- Persistent work. A team set up on Monday is still there on Friday: memory, folders, models and schedules survive between turns.
- Local control. Run it on your machine or a server you own, with your own model keys. Nothing is a hosted account.
- Bounded autonomy. A running turn can be stopped; approvals gate sensitive actions; merge authority stays with people.
- Composable plugins. Tree, social, timeline, workflow and project are separate plugins over the same
ds-botcore.
Install
npx -y @deepseek-ai/dsh@0.2.1-rc.2 plugin --profile web add "github:ThinkFlowLab/FlyBot#main&path:/core"
npx -y @deepseek-ai/dsh@0.2.1-rc.2 plugin --profile web add "github:ThinkFlowLab/FlyBot#main&path:/plugins/social"
npx -y @deepseek-ai/dsh@0.2.1-rc.2 plugin --profile web add "github:ThinkFlowLab/FlyBot#main&path:/plugins/tree"
The first line is the FlyBot core; the other two turn it into a team. Add workflow, timeline or project the same way. For an always-on instance, use the Docker deployment in Getting started.
What it is not
- Not a single chatbot: it is a team with a visible tree and roles.
- Not a hosted service: you run it and supply the model keys.
- Not a framework for building agents: it is an app on top of DeepSeek Harness.
- Team authority across multiple human owners is planned, not shipped; the current team mode stays disabled until its authorization gates pass.
Next
- Getting started: install, run the first team, and deploy with Docker.
- Architecture: core, patches, plugins and subsystems.
- Sandbox: file policies, approvals and preflight.
- Development: code standards, tests and the contribution loop.