ABOUT

Who we are.

A small team building the thing we wish we had.

Carabase is built by a small, distributed founding team. We are engineers and operators who spent the last few years watching friends, colleagues, and ourselves give up on the promise of personal AI — and we got tired of waiting for someone else to fix it.

What we're building

A personal context engine. Self-hosted. Sub-10ms retrieval. A persistent, temporal, queryable model of everything you know, who you know it from, when you learned it, and how much you trust it. Three surfaces — Host, Desktop, iOS — running on hardware you own.

It is not a note-taking app. It is not a chatbot. It is not a wrapper on someone else's model. It is the missing substrate — the thing every useful AI agent should have been talking to from the start.

The full argument is in the manifesto.

Why collective authorship

Most product pages lead with a founder photo and a bio. We are not there yet, and that is deliberate. Carabase is pre-launch. The thing that matters right now is the thesis and the software, not the résumé.

When we leave early access, the bylines come on. Until then, everything on this site is published under a collective voice: The Carabase Team. Our code — and the commits that produced it — is public on GitHub. If you want to know who is doing the work, the git log is the authoritative answer.

Principles

Self-hosted or nothing. Your context is the most sensitive data you own. It will never touch a server we control. Carabase Host runs on your hardware; your knowledge graph is a single Postgres database you can shut down, back up, or delete with one command.

Agent-native. We ship an MCP server on day one. Any AI tool that speaks the Model Context Protocol can plug into your context without a bespoke integration. We are building the substrate, not the chat window.

Speed is a feature. Sub-10ms retrieval is not a nice-to-have. It is what separates a tool you reach for from a tool you route around.

Open where we can be. The ingestion adapters, the MCP server, and the client shells are open source under the OpenClaw ecosystem. The knowledge graph itself is yours — there is no proprietary format, no lock-in, no "premium tier" gate.

Getting in touch

Email us at hello@carabase.ai. If you are a founder or operator interested in early access, that is also where to write. The waitlist lives at the bottom of the homepage.

For security reports, see /.well-known/security.txt.

Imagine the answer arriving in the same breath as the question.

Request early access →