nanobot is a self-hosted personal agent runtime written in Python that runs in a browser UI, a terminal client, or a chat app. A small agent loop sits at the centre: messages arrive from a channel, the model decides when a tool is needed, and memory or skills are pulled in as context rather than through a separate orchestration layer. Around that core it bundles tools, long-term memory, MCP integrations, model routing, multi-agent delegation, scheduled automation, and an OpenAI-compatible API.

Built-in tools cover files, shell, web search, web fetch, MCP servers, cron, image generation, and subagents. Session history and long-term memory persist through a component called Dream, so long-horizon goals and scheduled automations survive across runs. Chat channels include Telegram, Discord, Slack, WeChat, Feishu, Microsoft Teams, email, and Mattermost, and a Python SDK plus an OpenAI-compatible API let other software drive the same runtime.

Processes are organised around a shared local gateway. nanobot webui creates the config and workspace, starts or joins the gateway, and opens a localhost UI that is not exposed to the LAN; nanobot gateway --background promotes that gateway to a persistent background service so channels and automations keep running after every local client exits, with status, logs, restart, and stop subcommands. Running nanobot alone opens the terminal client in the launch directory, sharing conversations and the gateway with the browser UI, and nanobot -m "..." performs one request and exits.

The web UI ships inside the published wheel with no separate frontend build. It keeps separate topics per task, offers temporary chats that are not written to history or memory, shows reasoning, tool calls, file edits, diffs, command output, and generated artifacts, and lets you switch models and workspaces mid-conversation as well as manage providers, channels, Apps, Skills, and Automations.

Installation needs Python 3.11 or newer, through a one-command installer, uv tool install, pip, or an editable Git checkout (which additionally needs Bun to run the matching terminal client). Deployment options include a one-click Render blueprint that asks for an Anthropic API key and a web token and provisions persistent storage, plus Docker, Docker Compose, Linux services, and a macOS LaunchAgent. Release v0.3.0 added inline subagents consulted without leaving the current task, per-session model presets switchable from the composer, guided first-run setup, and live configuration reloads.

Features

Three front ends: a bundled browser workbench, a native terminal client, and chat apps, all sharing one local gateway and conversation history
Chat channels: Telegram, Discord, Slack, WeChat, Feishu, Microsoft Teams, email, and Mattermost
Built-in tools: files, shell, web search, web fetch, MCP servers, cron, image generation, and subagents
Memory: session history and long-term memory through Dream, keeping context across long-running work
Automations: scheduled and triggered agent work, with long-horizon goals that run over many steps
Model routing: OpenAI-compatible providers, local models, fallback chains, and per-session model presets
Multi-agent delegation: inline subagents consulted without leaving the current task
Integration surfaces: a Python SDK and an OpenAI-compatible API for driving the runtime from other software
Background gateway: a persistent service that keeps channels and automations alive after local clients exit
Deployment: a Render blueprint, Docker and Docker Compose, Linux services, and a macOS LaunchAgent
Localisation: documentation in ten languages, with workspace and access-mode controls per topic