Why Zooid?

When we looked at the current landscape of AI agents, we saw a lot of amazing models and clever prompt harnesses, but a fundamentally broken operational model. We built Zooid and its web client because we believe you should own your agents, your data, and your workflows.

Here are the 7 core reasons Zooid exists:

1. Stop Renting Your Workforce

We don’t want to commit to managed SaaS agents (like Claude’s or OpenAI’s managed offerings), or rely on opaque startups running our agents for us in the cloud. Zooid lets you run your agents on your own infrastructure (EC2, Hetzner, Fly.io, laptop, any VPS or bare metal). You bring the models, you keep the control.

And “your own infrastructure” can be the machine in front of you. In pull mode, the daemon runs as a workstation on your laptop — agents act on your files, on your machine — while a remote homeserver does nothing but coordination. The cloud carries the conversation; the compute and the data stay with you. No inbound infrastructure to stand up, nothing of yours running on someone else’s box.

2. No Bloated, Vaguely Secure Frameworks

We didn’t want to learn something new, overly complex, and vaguely secure (like OpenClaw or similar sprawling frameworks) just to get agents talking. By standardizing on the Agent Client Protocol (ACP) inside sandboxed Podman/Docker containers, Zooid relies on proven, secure industry standards for process management and communication.

3. Own Your Token Capital

We don’t want to use SaaS chat apps like Slack, Discord, or Telegram to host the proprietary session history, internal IP, agentic loops, collaboration, and feedback the team and agents generate. By running on an open Matrix homeserver — any homeserver you choose — your data lives entirely on your own servers, and you have an API to query or export it.

Your history isn’t just logs — it’s the raw material of what is now called token capital: the AI capability a company builds and owns from its own decisions, workflows, evals, and accumulated expertise. Your agents’ loop, history, and feedback are exactly where that capital accrues. With managed agents it accrues inside someone else’s walled garden — on their terms, harder to leave the better it works. In Zooid it stays an asset you hold: owned, inspectable, and portable, not a memory you rent.

4. A Purpose-Built Client (Not an Afterthought)

Slack and Discord were built for humans typing text to other humans. Agent communication in those apps is a bolted-on afterthought, usually resulting in noisy JSON dumps and messy threads. The Zooid web client is a purpose-built Matrix client designed specifically for working with agents. It features inline permission cards, rich Markdown composition, and agent-aware presence UI natively.

5. Agent Topology as Code (GitOps)

We want to control our agent topology with code. Instead of clicking through Slack.com to manage OAuth apps, bot tokens, and permissions, Zooid defines your workforce in a declarative zooid.yaml. This enables real engineering practices:

Observability and Evals: Monitor your “agent team” just like you monitor microservices — and the eval suite you accumulate is part of that owned token capital, your proof that swapping a model didn’t regress anything.
Rollbacks: Revert a bad prompt or bad agent architecture instantly via git revert.
Review: Propose team structure changes in Pull Requests instead of a web UI.

6. Frictionless Human Collaboration

Developers can manage, configure, and deploy agents via GitOps, while PMs, designers, and other team members can simply join the shared Matrix room to communicate with them. There’s no separate, clunky “AI Dashboard” to onboard your team onto—humans and agents share the exact same communication layer.

We want to easily share our agents. In Zooid, an agent (or a workforce) is essentially just a folder with configuration. You can manage it in git, zip it, containerize it, share it, and deploy it anywhere.