Archē (ἀρχή) - /ˈɑːrkiː/ - ancient Greek for first principle; the origin from which everything else is derived.
Most teams' AI tooling today has the same problem: it's a pile of disconnected things. A chatbot here, a knowledge base there, some integrations duct-taped together. Every team has their own setup. Nothing talks to anything else. And the moment you ask "how did we fixed this issue last week?", nobody knows.
And here's a subtler problem: some of your teammates are getting dramatically better results from the exact same LLMs you're both using. Why? Because they have a better prompt. They've figured out the right context to feed in. But that knowledge lives in their head, maybe in a Notion doc nobody reads, and it never becomes something the whole team can rely on.
We've been thinking about this for a while at Peaberry Studio. And this week, we shipped our answer: Archē.
What Archē is
Archē is a self-hosted agentic operating system for teams. The idea is simple but the execution matters: instead of a single generic chatbot, you get a coordinated team of specialized agents, each one configured with specific roles, tools, and knowledge relevant to how your team actually works.
The core pieces:
Agents: not one assistant but many, each scoped to a job. Support, copywriting, engineering, requirements. You configure the model, the tools, the permissions. They know your products, your tone, your processes.
Living knowledge base: documentation that doesn't rot. Version-controlled, inline-editable, with diff reviews and full traceability. When an agent uses a document, it uses the current version. When your process changes, the knowledge changes too.
Connectors: Linear, Notion, Jira, and custom MCP endpoints. Real-time status, connection testing. Your agents aren't isolated; they know what's happening in your tools.
And all of this deploys with a single CLI command, on your own infra (AWS, GCP, Azure, or on-prem). Nothing leaves your network unless you want it to.
Why self-hosted
Because "trust us with your data" isn't a strategy anymore.
With Archē, your institutional knowledge stays where it belongs: inside your infrastructure. The agents run there. The knowledge base lives there. The audit trail is yours.
This isn't just a security argument. It's a control argument. If the model changes, you decide when to upgrade. If the integrations break, you fix them. You own the system.
Open source
The full source is on GitHub. Explore it, break it, extend it, or just read it. We built this in the open because that's how we like to work.
Iñaki already announced it on X, so go take a look as he also uploaded a bunch of videos explaining how it works.
We're still early. There's a lot more coming. But the foundation is solid, and I'm proud of what we shipped. If you're building for teams that are serious about AI infrastructure, and serious about not depending on someone else's cloud for it, this is worth your time.