nanobot

About nanobot

Self-hosted personal AI agent framework in Python — a terminal agent and a local web UI, with built-in tools, persistent memory, scheduled automations, and first-class Ollama support so the whole loop can run on your own hardware.

Best for: Developers who want a persistent personal agent — one that remembers, runs on a schedule, and can execute tools — without renting it from a cloud service or wiring the orchestration themselves.

Deployment: Install with pipx (pipx install nanobot-ai) or uv. Requires Python 3.11+. On Ubuntu 24.04+ and other PEP 668 distributions, a plain pip install into system Python is blocked by design — use pipx. Start the web UI with 'nanobot webui'; it binds to 127.0.0.1:8765 and is not exposed to the LAN without an explicit host change plus a password.

Skill level: Intermediate

Tradeoffs

It is an agent with shell and file-write tools pointed at your machine, so run it against a workspace you are willing to let it modify. A model provider must be configured before first use — it ships with none selected. Some bundled skills depend on external CLIs and report themselves unavailable until those are installed. The project is young, so expect API and configuration churn.

Related guides and resources

Explore step-by-step setup guides, comparisons, and stack recipes for this tool category.

Best for

Developers who want a persistent personal agent — one that remembers, runs on a schedule, and can execute tools — without renting it from a cloud service or wiring the orchestration themselves.

Why use it

Most agent frameworks are libraries you build with; nanobot is an agent you run. Installing it gives you a working agent immediately — web UI on localhost, a terminal mode, a memory store, a cron scheduler, and around nineteen registered tools covering file editing, shell execution, and web fetch. Local models are a first-class provider rather than an afterthought: its provider registry carries an explicit local flag and ships a default Ollama endpoint at http://localhost:11434/v1, alongside vLLM, LM Studio and OpenVINO Model Server. MIT licensed.

Key features

• Runs as a persistent agent, not a library: web UI, terminal mode, and a background gateway process
• First-class local providers — Ollama, vLLM, LM Studio and OpenVINO Model Server are flagged as local in the provider registry, with Ollama defaulting to localhost:11434
• Persistent memory plus a git-backed workspace, so context survives restarts
• Built-in cron scheduling and a heartbeat loop for recurring autonomous tasks
• Installable skills, with availability checks that surface missing CLI dependencies rather than failing silently

Common AI use cases

• Run a private personal assistant that keeps context between sessions
• Schedule recurring agent tasks such as digests or repository checks
• Give a local model real tool access — file editing, shell, and web fetch — behind a local web UI

Who should use it

• Developers who want an agent that persists and runs on a schedule rather than a stateless chat window
• Self-hosters keeping assistant data and model inference on their own hardware
• Anyone prototyping agent workflows against Ollama without building orchestration first

Who should not use it

• Anyone unwilling to grant an agent shell and file-write access to a working directory
• Teams needing a mature, stable API — the project is early and moving quickly
• Users who want a plain chat interface with no agent behaviour — a chat client is a better fit

Tradeoffs

It is an agent with shell and file-write tools pointed at your machine, so run it against a workspace you are willing to let it modify. A model provider must be configured before first use — it ships with none selected. Some bundled skills depend on external CLIs and report themselves unavailable until those are installed. The project is young, so expect API and configuration churn.

Alternatives

• Open Multi-Agent
• Flowise