Home / Directory / Personal agents
Alternatives to hosted AI assistants
Personal agents you run yourself, that live in your chat apps and act with your permission.
Ranked by score out of 30, most convivial first.
nanobot
A small self-hosted Python assistant that works through a web page, a terminal or your chat apps.
IronClaw
A Rust assistant that runs untrusted tools in WebAssembly sandboxes and says it collects no telemetry.
Moltis
A single Rust program that runs a persistent assistant on your own hardware, with every command sandboxed.
OpenClaw
An open-source AI assistant that runs on your own computer and works through your chat apps.
Hermes Agent
An open-source AI agent that learns skills from the tasks it does and remembers you.
NanoClaw
A small assistant that runs each agent in its own Linux container and keeps credentials outside it.
About these tools
A personal agent waits in the background, answers in the chat apps you already use, remembers things and acts on your behalf, so you want it running on a machine you control. OpenClaw started this style of assistant, Hermes Agent learns skills from the tasks it does, and nanobot is a small Python version with local model support. NanoClaw runs each agent in a container, IronClaw sandboxes its tools in WebAssembly, and Moltis is a single Rust program.
They act with your files and logins, so keep their access narrow, run them in a container where you can, and treat incoming messages as untrusted.
How to choose
Start with how much setup you can take. nanobot is the gentlest, OpenClaw and Hermes Agent have the biggest following, and NanoClaw, IronClaw and Moltis are for people who want sandboxing first. Then check which model provider each needs, since a few default to a hosted one.
Frequently asked questions
Is there an open-source alternative to hosted AI assistants?
Yes. OpenClaw, Hermes Agent, nanobot, NanoClaw, IronClaw and Moltis are open source agents you run on your own computer or server.
Are personal agents safe to run?
They can be, with care. They act with your logins, so keep their access narrow, run them in a container where you can, and treat incoming messages as untrusted. NanoClaw, IronClaw and Moltis are built around sandboxing.
Do I still need an AI model provider?
Not always. nanobot supports local models through Ollama and vLLM, and Hermes Agent and OpenClaw can use models you choose. IronClaw asks for a provider API key during setup.