Open-source alternatives to hosted AI assistants
AI agents you run yourself, on models you choose.
An AI agent doesn't just answer questions. It acts for you, which means it needs your logins, your files and your trust. Hosted assistants keep all of that on someone else's servers. OpenClaw and {hermes-agent} are open-source agents you run on your own machine, working through the chat apps you already use and connecting to whichever AI models you pick.
Both are powerful and both are young, so we score them on how much control they give you and mark how much care they ask for.
The tools
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.
A benchmark review of the biggest tool in this category is on its way, so you can see what these are measured against.
How to choose
Start with what you are willing to hand over. An agent that can send email as you can also make mistakes as you, so give it as little as the job needs, and read the project's security guide before you connect anything important. Then look at who steers the project: a non-profit foundation and a venture-backed company are different kinds of keeper, even when both use the MIT licence.
Questions people ask
Is there an open-source alternative to ChatGPT agents?
Yes. OpenClaw and {hermes-agent} are both MIT licensed and run on your own computer or server.
Are AI 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. OpenClaw's own guide says so.
Do I still need an AI model provider?
Usually yes. The agents are free, but they send your prompts to a model you choose, either a hosted one you pay for or one you run yourself.