tinyhumansai/openhuman

OpenHuman – Fast, cheap, and scalable AI‑agent harness

What it is – OpenHuman is an open‑source, Rust‑based agent harness that lets you run hundreds of AI agents on a tiny server (e.g., 500 agents on a $10 VPS). It ships both as a desktop application for end‑users and as a Rust library that developers can embed in their own code.

Key selling points

Feature	Why it matters
Speed – cold‑start in ~0.1 s, typical task finishes in ~20 s (fastest of seven harnesses tested)	Reduces latency for interactive agents.
Cost efficiency – 2.6× fewer tokens per task, lowest total bill in benchmark	Cuts API usage fees, especially when scaling to many agents.
Resource efficiency – ~68 MB peak memory, ~1.3 s CPU time per task (≈8× less than rivals)	Allows dozens‑to‑hundreds of agents on cheap VPS hardware.
Rust‑first library – single‑process compiled core, callable like any Rust function	No extra daemon, easy to embed in Rust services, supports fleet deployment.
Built‑in memory & token compression – RLM‑based compression, selective recall with citations	Keeps prompts small while preserving context, improving speed and cost.
Instant tool search (Jev) – tiny model finds the right tool among >1,200 options with 86.8 % top‑pick accuracy	Faster, more reliable tool selection than keyword search.
Plug‑in architecture – unified Rust bus for features (search, browser, voice, etc.)	Load only what you need; a stuck plug‑in doesn’t crash the whole system.
Rich integrations – 119 OAuth apps via Composio, MCP/skill bundles, 14 messaging channels, real‑browser & desktop control, voice agents, workflow graphs, safety gates, local‑model support (Ollama, LM Studio, MLX)	Provides a full‑stack agent platform out of the box.

How to get started

• End‑users: download the desktop app (macOS .dmg, Windows .msi/.exe, Linux .deb/.AppImage) from the website or the latest GitHub release.
• CLI lovers: run the provided install script (curl … | bash on macOS/Linux, PowerShell on Windows) which picks the right package, verifies checksum, and installs it.
• Developers: add the openhuman crate to a Rust project and build an agent with a few lines of code (see README snippet). The library includes a typed API, embedding guide, and examples.

Typical use cases

• Customer‑facing SaaS that needs a dedicated agent per user – you can spin up hundreds of agents on a single cheap VPS instead of provisioning a separate container per user.
• Automation workflows – combine browser control, shell tools, and custom memory engines to build end‑to‑end bots (e.g., code reviewers, data extractors, voice assistants).
• Research & benchmarking – the public openhuman‑benchmarks repo lets you reproduce the SWE‑bench performance numbers and compare against other harnesses.
• Local‑only privacy‑focused agents – run with local models (Ollama, MLX) and enable Privacy Mode for zero‑API‑call operation.

License & community

• Licensed under GPL‑3.0.
• Early‑beta status – active development, public issue tracker, Discord, Reddit, and GitHub Discussions for support.
• Contribution guide and CI checks (Rust cargo, Node pnpm) are provided for both UI and core development.

---

Bottom line – OpenHuman is a genuine, production‑grade AI‑agent harness that focuses on speed, cost, and scalability through a Rust‑centric design. It’s suitable for both power users who want a desktop AI assistant and developers who need an embeddable, fleet‑ready agent engine.