NanoClaw is open source software for building your own personal AI agent that runs on your own computer or server. It is developed by NanoCo, based in Israel, and was created as a lightweight and secure alternative to the earlier OpenClaw project. The official tagline is “Your personal AI agent. Secure. Lightweight. Yours.” Its defining trait is that agent execution is confined inside a Docker container, so the agent can only see what you explicitly mount. You talk to it through the messaging apps you already use, such as WhatsApp, Telegram, Slack, and Discord. It is released under the MIT license, so the software itself costs nothing to use.
Key Features
- Safe execution through container isolation: Each session starts an unprivileged Docker container, and the agent can only reach files that were explicitly mounted. Rather than relying on checks inside the application, NanoClaw draws the boundary at the OS level
- Talk to it from a messaging app: It supports many channels including WhatsApp, Telegram, Slack, Discord, Teams, Signal, Matrix, and iMessage. You do not have to open a dedicated interface ─ you give instructions in the same conversation flow you already use
- A “Skills” approach that adds only what you need: Instead of bundling everything up front, you pull in only the channels and integration modules you want with commands such as
/add-<channel>. The core stays small and remains readable - A separate workspace per agent: Instructions, memory, credentials, and mounted files can be kept separate per agent, which makes it easy to run one agent for work and another for personal use
- Credentials are never handed to the agent: API keys and similar credentials are held in the Agent Vault, so the agent itself never holds the raw keys
- Scheduled tasks: You can register recurring jobs so that work runs at set times
- Being lightweight is the point: The official comparison lists 200 TypeScript files and roughly 29,300 lines of runtime code, set against OpenClaw’s 3,680 files and 434,453 lines
Pricing
| Plan | Price | What it covers |
|---|---|---|
| NanoClaw itself | $0 (MIT license) | Every feature is available as open source |
| AI model usage | Depends on each provider’s billing | Claude, OpenAI and others require an API key or a subscription. Running a local Ollama model adds no extra cost |
Pricing is current as of August 2026. Please check the official site for the latest pricing.
NanoClaw itself is free, but running it in practice means paying for the AI model behind it. If you choose a local model, you can run it with essentially no additional cost.
Pros & Cons
✅ Pros
- Because container isolation is the default, the risk of accidents when you hand real work to an agent is structurally reduced
- The codebase is small enough to follow on your own, and the project is designed on the assumption that you will fork it and make changes
- You use it from the messaging apps you already have, so there is little new tooling to learn
- Fully open source under the MIT license, with no vendor lock-in
- You can choose your model provider (Claude, OpenAI, a local Ollama model, and others)
⚠️ Cons
- You have to install and run it in your own environment, which means preparing Docker, Node.js, and pnpm. This is not a no-code SaaS product
- Docker is the only supported container runtime, and running on Windows requires WSL2
- Customization is designed around handing the work to an AI coding agent such as Claude Code, which can be confusing if you are not used to that workflow
- Support comes mainly from the official Discord and the documentation, so it does not suit operations that assume a commercial support desk
- Model usage costs fall on you, and depending on how you use it those costs can be hard to predict
Comparison with Similar Services
| Item | NanoClaw | OpenClaw | Claude Code | n8n |
|---|---|---|---|---|
| Positioning | Personally owned AI agent platform | The earlier project in the same category | Coding agent that runs in the terminal | Workflow automation tool |
| Main entry point | Messaging apps (WhatsApp, Telegram, etc.) | Messaging apps | Terminal | Web admin interface |
| Isolation model | A Docker container per session | Mostly application-layer controls | Local execution with permission prompts | Workflow execution environment |
| License | MIT (open source) | Open source | Commercial (included with a plan) | Source-available |
| Setup effort | Self-hosting required | Self-hosting required | Just install the CLI | Self-hosted or cloud version |
Who Is It For
- People who want a personal assistant they can message from a chat app, running on their own machine instead of someone else’s server
- People who want an agent to handle files and run commands, but feel uneasy about doing so in an environment with no isolation
- Developers who want to read the code, understand it, and adapt it to their own needs
- People who want separate agents for work and personal use, each with its own memory and permissions
- People who want to keep running costs down by using local models
Summary
NanoClaw is an open source project aimed at one thing: running a personal AI agent under your own control, in an isolated environment. Its container-first design and its Skills approach of adding only what you need are both meant to structurally reduce the anxiety of handing real work to an agent. Setting it up requires Docker and Node.js, so it is not for everyone, but it is a strong option if you want an AI agent that stays entirely on your own machine. A good first step is to check the official documentation and the GitHub repository to see whether it fits your environment.