OpenClaw didn’t just get popular — it got rewritten, repeatedly, within weeks. Four notable forks and rewrites appeared, each taking the core idea (a self-hosted, model-agnostic assistant) and trading away some of OpenClaw’s bulk for a specific advantage: a smaller codebase, lower hardware needs, a single binary, or container isolation. The result is a small family of projects that look similar but fit different situations.
This comparison lays out OpenClaw and its four main relatives — nanobot, ZeroClaw, PicoClaw, and NanoClaw — what each one trades, and how to pick. If you just want the original explained, see what OpenClaw is.
The family at a glance
OpenClaw and its main forks
| Project | Language |
|---|---|
| OpenClaw | TypeScript |
| nanobot | Python |
| ZeroClaw | Rust |
| PicoClaw | Go |
| NanoClaw | TypeScript |
The pattern: OpenClaw is the maximal version, and each fork picks one axis — simplicity, resource use, portability, isolation — and optimizes hard for it while shedding everything that gets in the way.
Source links
| Project | Source | Why open it |
|---|---|---|
| OpenClaw | openclaw/openclaw | Start here if you want the full feature set and the original docs. |
| nanobot | HKUDS/nanobot | Check the Python rewrite and its current install/config instructions. |
| ZeroClaw | zeroclaw-labs/zeroclaw | Check the Rust rewrite if you care about small deployments and low resource use. |
| PicoClaw | sipeed/picoclaw | Check the Go/single-binary route for small-board or embedded-style setups. |
| NanoClaw | nanocoai/nanoclaw | Check the container-isolated approach if sandboxing is the main reason you are looking past OpenClaw. |
OpenClaw: the full experience
The original is the most feature-complete. It supports the widest set of messaging channels (Telegram, Discord, Slack, WhatsApp and more), voice, a companion app, the deepest skill ecosystem, and the most polish. It’s also the heaviest — it wants the most memory and has the largest codebase.
Choose OpenClaw when you want everything and have a normal machine or VPS to run it on. If your needs are mainstream — a capable assistant on a laptop or a modest server — there’s no reason to reach for a fork. The install guide gets you going, and the source repo is openclaw/openclaw.
nanobot: minimal and auditable
nanobot is the Python rewrite for people who want to understand their assistant. The selling point is that the codebase is small enough to read in an afternoon, which matters if you’re security-conscious or want to modify it without spelunking through a large project.
You give up breadth — fewer channels, fewer built-in skills — for a codebase you can hold in your head. Good for tinkerers, auditors, and anyone who distrusts running a large opaque process with file access. Start from HKUDS/nanobot for the current setup notes.
ZeroClaw: tiny hardware and VPSes
ZeroClaw is the Rust rewrite built to run where OpenClaw would struggle — cheap VPSes, low-memory boxes, the kind of $10/month server people leave running 24/7. Rust’s efficiency means it sips resources, which is exactly what you want for an always-on assistant you don’t want to babysit or overpay for.
Reach for ZeroClaw when you’re putting an assistant on a small always-on server and OpenClaw’s footprint feels heavy. Start from zeroclaw-labs/zeroclaw and check its current provider examples before assuming an OpenClaw config transfers.
PicoClaw: a single binary on a Raspberry Pi
PicoClaw takes “lightweight” to the extreme. Written in Go, it ships as a single binary with a tiny memory footprint and near-instant startup, designed to run on cheap single-board computers — a Raspberry Pi, or even smaller. No runtime to install, no dependency tree; just a binary you drop and run.
It’s the pick for a dedicated little always-on device: a Pi in a drawer running your assistant, sipping power. You trade features and channels for portability and minimalism. Start from sipeed/picoclaw.
NanoClaw: container-isolated safety
NanoClaw is the TypeScript fork that leans into security through container isolation — the assistant runs inside containers, so its file and command access is sandboxed rather than running loose on the host. For an agent that executes commands, that boundary is reassuring. Its tagline is “fork it, ship your own,” so it’s also a base for people building their own assistant.
Choose NanoClaw when you want the agent’s reach contained by design, or you’re building a derivative. Start from nanocoai/nanoclaw and read its current container setup before running it with file access.
How to choose
A decision path:
Which one fits
| Want the most features and channels | OpenClaw |
|---|---|
| Want a small codebase you can read and trust | nanobot |
| Running on a cheap VPS or low-memory box | ZeroClaw |
| Running on a Raspberry Pi / single-board computer | PicoClaw |
| Want the agent sandboxed in containers | NanoClaw |
They share the model-agnostic DNA
One thing carries across the whole family: you bring your own model. Whether you run OpenClaw, ZeroClaw, or PicoClaw, you can point it at a cheap option like DeepSeek or GLM, or a local model. The provider config differs per fork, but the freedom to choose the brain — and keep costs low — is common to all of them.
Picking between OpenClaw and a fork
- Default to OpenClaw unless a fork's advantage clearly fits you
- nanobot for a readable, auditable codebase
- ZeroClaw for cheap VPSes and low-memory hosts
- PicoClaw for a Raspberry Pi or single-board computer
- NanoClaw for container isolation or building your own
- Follow each project's own docs — they're not interchangeable
Wrapping up
OpenClaw is the full-featured original; its forks each pick one thing to be great at. nanobot is the readable one, ZeroClaw the tiny-hardware one, PicoClaw the single-binary Pi one, NanoClaw the sandboxed one. Default to OpenClaw, and switch to a fork only when its specific advantage matches your hardware or your priorities — and remember they’re distinct projects with their own docs and skills.
For the original’s full capabilities, see the OpenClaw power-user guide.
