The orchestrator
zeno
The always-on core. Runs your company's processes and puts scoped, short-lived agents to work in one space your whole team shares.
Open source · Self-hosted
An open framework for
An open framework for
AI-native companies
Meet zeno — an always-on orchestrator that runs your company's processes and puts agents to work in one space your whole team shares, on infrastructure you own. Describe the work once; it gets done.
- One shared space
- Deterministic by default
- Cheaper the more it runs
- Self-hosted
- Open source
An AI-native company runs on agents. zeno makes them trustworthy — working in the open where your team can see them — and cheaper the more they run, on infrastructure you own.
A team can't run on personal agents
Today's AI agents are personal, per-tab, and forgetful. That's fine for one person — it breaks the moment a team tries to run real work on it.
The old way
Everyone drives their own agent in their own tab, then merges the results by hand. Work diverges, nothing carries over from one task to the next, security is bolted on afterward, and the whole thing lives on someone else's platform.
With zeno
One orchestrator runs your processes in a space the whole team shares. Agents work in that same context — reading the same discussion, writing to the same knowledge base — on infrastructure you own. Team agents, not personal ones. Nothing to merge.
One orchestrator at the center
zeno is the always-on core. It runs your company's processes and calls in an agent only for the parts that need judgement — each one given just the access its task requires, and nothing more.
The shared space
Your team and its agents in one place, over the channels you already
use — chat, Telegram, GitHub issues, email.
↕ messages in · answers out
zeno — the orchestrator
Runs your processes, spawns short-lived, scoped agents for the
judgement calls, and asks the team when it needs a decision — picking
up right where it left off when the reply lands.
↑ helped by
What helps you run it
nixops stands zeno up on servers you own, and
machines package ready-made processes you drop in to
put it to work.
One framework, any company
zeno isn't tied to one use case. Describe any company's processes — support, operations, research, back-office, publishing — as code, and it runs them with your agents, in your space, on your servers. The framework is built and running today; what you put on it is up to you.
Any process
Spell out a routine once, in plain deterministic code. zeno runs it exactly the same way every time and calls an agent in only where judgement is needed.
Any agents
zeno spawns short-lived agents for the judgement calls — each boxed off and scoped to exactly what its task needs, holding no standing access of its own.
Any channel
Your team works with those agents wherever they already are — chat, Telegram, GitHub issues, email — and the work waits for a human whenever it should.
We build on it ourselves — this site's own publishing and research run on zeno, end to end — but that's one example, not the point. See how it works →
It gets cheaper the more it runs
Most AI agents are goldfish — they solve every task from scratch and forget it the moment it's done.
Goldfish
Do the same job a thousand times and you pay full price a thousand times. Nothing carries over. The cost never comes down.
A company that remembers
zeno turns work it has already figured out into reusable process. The first run is the expensive one; after that, repeats run on their own — cheaper and faster every time. Experience becomes an asset that grows with use.
Cost-per-task falling as volume grows is the bet we're building toward — proven out one working workflow at a time, not a number we're quoting.
Why it's different
Three things most agent tools don't give a team at once.
Not a personal tab
Personal agent apps give each person a private assistant. zeno is one shared space for the whole team, so work doesn't diverge and there's nothing to merge by hand.
Not per-task glue
Other tools wire up each workflow by hand, every time. zeno's processes are deterministic and compounding, and every agent is scoped to just what its task needs.
Not a closed platform
No service in the middle. zeno is open source and self-hosted — your processes, your data, your servers.
Principles
The beliefs the whole stack is built on.
One shared context
Work happens in a space the whole team sees, not in private tabs. Agents read and write the same context, so there is nothing to merge.
Deterministic by default
Spell out the routine once and it runs the same way every time. An agent is called in only where something genuinely needs judgement — not for everything.
Security by design
Security is built into the structure, not bolted on. Each agent gets only the access its task needs and nothing more — secrets stay with the orchestrator, so a stray instruction can't reach what was never handed over.
Meet humans where they are
Communication is separated from work. An agent asks the team over whatever channel you already use — chat, Telegram, GitHub issues, email — and resumes when the answer lands.
Own your stack
Self-hosted from the metal up. Your processes, your data, your servers — no platform in the middle.
Reuse, don't reinvent
Best-in-class harness, mature communication software, proven infrastructure. We build the connective tissue and the boundaries between them — not worse copies of what already works.
Build a company that runs itself.
Start with the orchestrator and grow the stack from there.