Build your own AI assistant.
With safety and control.
Open-source. Runs on your machine. Asks before it acts.
Why a kit
OpenBotKit is a kit - not a service, not a platform. A kit gives you the parts and the tools to build something yourself. You own it. You control it. You don't depend on anyone else to keep it running.
We build kits because we believe people should own their software. The world has too many products that create dependency. Kits create independence.
We're living through a moment where, for the first time, people can genuinely own software that works for them. A personal assistant that reads your email, manages your calendar, and talks to your team used to require a startup and a team of engineers. Now it requires a good harness around good tools. That's what OpenBotKit is. A harness. And harnesses need to be built right.
Why safety
The AI assistants we saw shipping open-source had no safety guardrails. Tools that could send emails, post messages, and modify calendars without asking. We think that's irresponsible, regardless of how good the code is.
We take inspiration from how Anthropic approaches safety - with deep respect for it at every layer, from their models to Claude Code. We want to bring that same care to open-source personal assistants. Not because we have to, but because it's the right way to build.
Being open source doesn't exempt us from building responsibly. If anything, it raises the bar. When anyone can run your code, it better be safe.
What we believe
- People should control their own software.
- AI makes it possible for people to own software that actually works for them - not software they work for. OpenBotKit is a kit you run on your machine, configure yourself, and extend how you want. No vendor lock-in. No subscriptions that disappear.
- Safety is non-negotiable.
- Every sensitive action - sending an email, posting to Slack, modifying your calendar - requires your explicit approval. This is enforced in code, not in prompts. The AI cannot skip it. We built 8 independent defense layers because one isn't enough.
- Your data stays yours.
- Everything syncs to SQLite on your machine. No cloud relay. No data harvesting. Your emails, messages, and notes never leave your device unless you explicitly send them.
- Transparency over magic.
- Skills are plain text files you can read in 30 seconds. Every tool call is logged. Every decision the assistant makes is auditable. No black boxes.
- Cost-conscious by design.
- AI tools shouldn't cost a fortune. We ship cost profiles so you know what you're spending before you spend it. You choose the LLM provider and model - Claude, Gemini, OpenAI, OpenRouter, or Groq - and stay in control of the bill.
- Reliability over features.
- We'd rather have five integrations that never break than fifty that sometimes work. Your assistant should work the same way every morning. We test everything end-to-end because code without tests has no value.
Get started
Single Go binary. Runs on macOS and Linux.