Quick Verdict
nanobot is an actively maintained, MIT-licensed personal AI agent runtime. Its stable release as of July 21, 2026 is v0.2.2. It is a good fit for technical users who want to own the runtime, keep history in their deployment, connect remote or local model providers, and combine tools, MCP, scheduled work, and chat channels behind one gateway.
Self-hosted does not automatically mean offline or safe. Remote model providers, search services, and chat channels receive the data required for their calls. File and shell tools can make real changes to the host. Users who prefer a mature managed coding product should compare Claude Code, Cursor, and GitHub Copilot. Choose nanobot when auditability and runtime ownership matter more than turnkey enterprise controls.
Best For
- Developers running a personal agent on their own computer or server
- Users combining remote APIs, local models, MCP servers, and chat apps
- Researchers who want a relatively small, readable agent loop
- Teams able to operate secrets, containers, permissions, logs, and backups
- Not ideal for users who cannot review commands or need vendor-provided SSO, policy, and compliance commitments
Key Features
- Self-hosted runtime: terminal, bundled WebUI, gateway, Python SDK, and an OpenAI-compatible API
- Local history and memory: sessions and workspace state persist in the deployment the user controls
- Provider choice: remote services, fallback routes, custom compatible endpoints, Ollama, and vLLM
- Real tools: files, shell, web search and fetch, MCP, cron automation, image generation, and subagents
- Chat channels: WebUI plus integrations such as Telegram, Discord, Slack, Feishu, WeChat, email, and Mattermost
- Readable architecture: a compact core loop surrounded by extensible providers, tools, memory, and channels
Use Cases
- Run a persistent personal assistant with history and long-term memory
- Schedule searches, summaries, and carefully scoped file workflows
- Compare remote providers or route selected work to a local model
- Use an agent as a front end for existing MCP servers
- Study and modify the message-to-context-to-tool execution loop
Pricing
nanobot is free under the MIT License. The current stable package is v0.2.2 and is distributed as nanobot-ai. Total cost depends on the deployment: hosted models, search, speech, image services, and cloud compute can charge separately. Local models need hardware and operations. The official Render blueprint requires paid persistent storage, so free source code should not be confused with free hosting.
Pros
- MIT licensed, self-hosted, inspectable, and modifiable
- Active v0.2.2 runtime with WebUI, SDK, API, channels, and automation
- Keeps history locally while allowing both remote and local model choices
- Broad MCP and tool support without turning the core into a large platform
Cons
- Shell, file, network, and MCP access increase prompt-injection and accidental-action risk
- Local history does not prevent remote providers or channels from receiving submitted data
- Enterprise governance and service commitments are not supplied as a managed product
- Users own patching, backups, uptime, key rotation, and cost controls
- Agent-generated commands remain fallible and should not run unsupervised on a high-value host
Alternatives
| Tool | Best for | Main difference | Tradeoff |
|---|---|---|---|
| Claude Code | Terminal coding | Mature managed coding-agent workflow | Relies on a commercial hosted service |
| Cursor | Developers wanting an AI IDE | Strong project context and multi-file editing | Not a self-hosted personal-agent runtime |
| GitHub Copilot | Mainstream IDE and enterprise users | Broad integrations and administration | Less runtime ownership and customization |
| LangChain | Teams building their own agent apps | Larger component and orchestration ecosystem | A development framework rather than a ready personal runtime |
FAQ
Is nanobot free?
The software is free under MIT, but model APIs, search services, cloud compute, and persistent storage may cost money.
Can nanobot run fully offline?
It can use a local model with remote integrations disabled. Verify every provider, search, channel, telemetry, and MCP setting before treating a deployment as offline.
Where is conversation history stored?
The self-hosted runtime stores workspace and session data in its deployment. The operator is responsible for disk permissions, retention, deletion, and backups.
Is shell access safe?
It is high impact. Use a dedicated low-privilege account or sandbox, restrict directories and commands, require review, and keep production secrets outside the environment.
Is v0.2.2 actively maintained?
Yes. It is the stable release at this page’s July 21, 2026 cutoff, and the project remained active. Pin versions and review release notes before upgrading production deployments.
nanobot or Claude Code?
Choose Claude Code for a managed coding workflow. Test nanobot when self-hosting, chat channels, persistent personal memory, and runtime modification are the priorities.
Bottom Line
nanobot packages model routing, local history, memory, tools, MCP, automation, and chat channels into a lightweight runtime the user can own. That control is its advantage and its operational burden. Start with non-sensitive data inside an isolated environment, then grant filesystem, shell, and external-system access gradually.