openhuman vs ZeroClaw
Two of the top projects in Assistants, side by side: who leads on each score part and who each one suits, then setup, license, activity and what each review found.
openhuman
Rust agent harness with desktop app, built-in memory and embeddable library
ZeroClaw
Single Rust binary agent runtime with 30+ channels and hardware access
Verdict
openhuman ranks higher, 84 to 79. Of the 5 score parts we measure for both, openhuman leads on 3, ZeroClaw on none and 2 are even.
- Adoptionopenhuman leads by 11 points, 77 to 66.
- FreshnessEven at 100.
- Maintenanceopenhuman leads by 11 points, 95 to 84.
- Easy to runEven at 67.
- Agent-readyopenhuman leads by 15 points, 100 to 85: a wide gap.
Pick openhuman if you are
Users wanting a self-run agent app, and Rust developers embedding agents
- Embeds as a typed Rust library, no sidecar or daemon needed
- Public benchmark repo; reports 68 MB peak memory and 19.8 s median task time
Pick ZeroClaw if you are
Tinkerers wanting an always-on agent on a Pi or home server with GPIO access
- Default supervised mode: medium-risk operations need approval, high-risk ones are blocked
- Hardware peripherals on Raspberry Pi, STM32, Arduino and ESP32 via a Peripheral trait
| What we compare | openhuman | ZeroClaw |
|---|---|---|
| Score parts, out of 100 | ||
| Adoption | 77, popular | 66, popular |
| Freshness | 100, active | 100, active |
| Maintenance | 95, healthy | 84, healthy |
| Easy to run | 67, easy | 67, easy |
| Agent-ready | 100, ready | 85, ready |
| Facts from GitHub and the README | ||
| Stars | 42.2k | 33k |
| License | GPL-3.0 (copyleft) | Apache-2.0 (permissive) |
| Last commit | Oct 2026 | Oct 2026 |
| Last release | Oct 2026 | Oct 2026 |
| Language | Rust | Rust |
| Docker | Yes | Yes |
| GPU | Optional | Not needed |
| arm64 or Apple Silicon | Mentioned | Mentioned |
openhuman
OpenHuman is a single-process Rust agent harness shipped as a desktop app (macOS, Windows, Linux), a terminal app, a headless server and an embeddable Rust crate. It bundles memory recalled with citations, browser and desktop control, MCP and skills, 14 messaging channels, durable workflows, and local or BYOK models. The README reports its own benchmark on ten SWE-bench tasks and claims 500 agents fit on a $10 VPS; it is labelled early beta.
Who it is for: Users wanting a self-run agent app, and Rust developers embedding agents
Strengths
- Embeds as a typed Rust library, no sidecar or daemon needed
- Public benchmark repo; reports 68 MB peak memory and 19.8 s median task time
- Local models (Ollama, LM Studio, MLX) or BYOK across 26 providers
- Approval gate, OS-jail or Docker sandboxing, secrets stored in OS keyring
Weaknesses
- Labelled early beta; README warns of rough edges
- Benchmark is self-run on only ten tasks
- GPL-3.0 license constrains embedding in proprietary products
- 119 OAuth app integrations go through Composio, an external service
- GPU optional
- Docker + Compose
- Needs Composio (OAuth integrations), Ollama / LM Studio / MLX (optional local models), Docker (optional sandbox)
- Models: Ollama, LM Studio, MLX, OpenAI-compatible APIs, 26 BYOK providers
ZeroClaw
ZeroClaw is one Rust binary that routes messages from 30+ channels (Discord, Telegram, Matrix, email, voice, webhooks, CLI) to an agent loop backed by Anthropic, OpenAI, Ollama or any OpenAI-compatible provider, with fallback chains. Tools cover shell, browser, HTTP, MCP servers and GPIO/I2C/SPI/USB on Raspberry Pi, STM32 and ESP32. Supervised autonomy, OS sandboxes and signed tool receipts gate each action.
Who it is for: Tinkerers wanting an always-on agent on a Pi or home server with GPIO access
Strengths
- Default supervised mode: medium-risk operations need approval, high-risk ones are blocked
- Hardware peripherals on Raspberry Pi, STM32, Arduino and ESP32 via a Peripheral trait
- HTTP/WebSocket gateway plus web dashboard for chat, memory, config and cron
- Dual-licensed MIT or Apache-2.0; installs as systemd, launchctl or Windows service
Weaknesses
- Hand-written TOML config; a minimal V3 config needs four sections before it runs
- README states no RAM figures and no gateway port
- Unix installer places the binary under the Cargo bin directory
- no GPU
- Docker + Compose
- Models: Anthropic, OpenAI, OpenAI Codex, Ollama, OpenAI-compatible endpoints
More in Assistants
Building with these?
We help teams pick an open-source AI stack and run it in production. Client work never changes a rank.