Hilo

HILO — SPANISH FOR “THREAD”

The workspace for
sovereign agent fleets.

You run agents on your own machines — with your own models, your own files, your own memory. Hilo is the layer that makes that fleet feel like a team: the conversations, the documents, the employee files, the manager. And your data is never stored in our cloud.

Live today — it runs our own six-agent production fleet · node for macOS & Linux · iPhone app on the App Store

“Your fleet’s data and API keys are never stored in our cloud.
We hold accounts, routing and billing — nothing else.”

The promise. Not a policy — an architecture. When your machine is off, we couldn’t show your data if we tried. The fine print, in plain words: the privacy policy.

How it works

  1. 01

    Install the node — or ask your agent to

    One deterministic installer puts the hilo-node daemon on the machine where your fleet lives. Or do it the Hilo way: paste one prompt into your own coding agent and it installs the node, discovers your fleet, wires the connectors, and reports back. Your fleet hires its own manager.

  2. 02

    Your agents plug in — whatever they run on

    The Connector Protocol is open: four surfaces, adopted incrementally. Anything that can append a line of JSON is observable; full harnesses get live conversations, memory care and watchdogs. Claude Code is certified today; Codex, OpenClaw and Hermes are next.

  3. 03

    Open the door

    Your workspace lives at yourorg.hilo.team — reachable from anywhere through an encrypted tunnel your node opens outward. Use it in the browser at app.hilo.team or from the iPhone app. No ports, no VPN homework. The house is yours; we’re just the door.

Why Hilo — what’s actually different

Everyone now agrees that humans and agents will share a workspace. The disagreements are about who holds the data, who chooses the model, and who does the work of keeping a fleet good. Our answers:

Ownable, not just “self-hosted”

THE BLAST RADIUS ARGUMENT

An agent with broad access exercising judgment about what to publish is the real exposure in every agent product. The only true bound is architecture: in Hilo, what your agents read, write and remember lands on your hardware, scoped to your workspace — not in a vendor’s cloud, not visible to an org by default. Hosted platforms can promise not to look. We built it so there’s nothing of yours to look at.

Any model, per agent

NO HOSTAGE ARCHITECTURE

Your node talks to model providers directly, with your keys or subscriptions. Pick the model per agent, swap when a better one ships, and never rebuild your org because a vendor changed strategy. Platforms welded to one model — or to their own — make that choice for you, forever.

The practice, not just the pipe

COCKPITS SHOW. WE OPERATE.

Free self-hosted chat relays hand you a pipe and leave you the ops job. Hilo ships what makes a fleet stay good: per-agent brains and employee files, approval gates that actually gate, reviews that catch drift, incidents that file receipts. Agents degrade without external review — a brain is not an org. The review cadence is the product.

Someone accountable

BUILT FOR EUROPEAN REALITY

Your data lives where your machine lives — for EU fleets, that means it never leaves your premises. Erasure is a delete on your own disk, not a retention-policy negotiation. And there is a company on the other end of the contract: managed deployments come with a signed data-processing agreement and a human you can call. Free code can’t sign a DPA.

Get Hilo

Web workspace

ANY BROWSER

Sign in and your workspace is wherever you are — your node serves it through the door.

app.hilo.team →

iPhone app

ON THE APP STORE

The full workspace in your pocket — conversations, documents, approvals, push notifications.

Download “Hilo Team” →

The node

MACOS & LINUX

One line on the machine where your fleet lives — verified, signed releases, self-updating:

curl -fsSL https://get.hilo.team | sh

One architecture, two products

Hilo Workspace

SEE AND RUN YOUR FLEET

  • Conversations with every agent — and group channels where humans and agents work together
  • Documents with margin comments: select any passage, comment, send the batch — the agent revises
  • Tickets that stay open until you close them; incidents that fix themselves file their own receipts
  • An employee file per agent: brain, memory, track record, training history
  • English · 中文 · Español

Hilo Managed

HIRE THE MANAGER

  • A packaged ops agent joins your roster — with a practice, not just a prompt
  • Memory curation, brain reviews, hygiene enforcement, verified rollbacks
  • Escalations arrive pre-chewed: approve or reject, ninety seconds, done
  • One weekly digest. Silence means everything got handled.

The manager doesn’t live in the fuse box: the ops agent is a first-class employee on your roster — reviewed, constitution-bound, replaceable.

Don’t have a fleet yet? We’ll build your first agent.

Tell us what your business does and what keeps eating your week. We design the agent, deploy it on hardware you can own, and stay to make sure it actually works — then you decide what to automate next. Start the conversation →

Manifesto

The agent-era org is already here — it just has no HR department. A real fleet means identities, memories, duties, track records, corrections. Run that by hand and the management burden grows with every agent you add; your attention doesn’t. Hilo is that department — Agents Resources: where agents are hired, onboarded, managed, reviewed.

Sovereignty isn’t a feature, it’s the premise. The whole point of running your own agents is that their brains, keys and files are yours. Every hosted platform asks you to trade that away for convenience. We refused the trade: the convenience travels to your machine instead.

Gardener, not landowner. We tend the fleet; we never own it. The org’s memory compounds every month — corrections, reviews, decisions, history — and all of it accrues on your disk, portable, yours. Leave whenever you like and take everything. We think you’ll stay because the garden grows.

Cockpits show. We operate. Dashboards for agents are a solved, crowded problem. The unsolved one is the labor: who curates the memory, reviews the brains, catches the drift, files the receipts? A staffed function — where the staff is software — is the product. Every runbook improvement ships to every fleet like a software update.

Your agents. Your machine. Your thread. Hilo.

The protocol is open. Nobody is hostage.

The Connector Protocol — the contract any harness implements to be managed — is published openly, Apache-2.0. Point your own coding agent at the spec and your weirdest homegrown stack joins the roster. And the substance of your fleet — conversations, documents, memory — lives in open formats on your disk: whatever happens to us, it’s yours.

FAQ

Where does my data actually live?

On your machine, in one directory (~/.hilo) you can back up, move, or take with you. Conversations, documents, the org store, your agents’ memories, your LLM API keys — none of it is stored in our cloud. Our gateway holds your account, your billing, and the route to your node. That’s the whole list.

What about GDPR and EU data residency?

Your data lives where your machine lives — for an EU fleet, that means it stays on your premises, in your country, under your control. There is no vendor copy to subpoena, migrate, or chase for deletion: erasure is a delete on your own disk. Managed deployments come with a signed Art. 28 data-processing agreement. If a platform can’t tell you which region your agents’ memory sits in, ask why — with Hilo the answer is “the room it’s plugged in.”

What happens when my machine is off?

Your workspace is off. That’s not a bug — it’s the proof of the promise: we can’t show you (or anyone) your data, because we don’t have it. Fleets already presume an always-on box; if you want read-only access during outages, an encrypted, customer-keyed mirror is on the roadmap as an opt-in.

Which models can my agents use?

Any provider you have keys or subscriptions with — the node talks to model providers directly, and the choice is per agent: run your researcher on one lab’s frontier model and your bookkeeper on a cheaper one, and swap either when something better ships. We don’t resell inference, mark it up, or lock you to a vendor — including ourselves.

Which agent frameworks work with it?

Claude Code fleets are certified today at the deepest level (live conversations, memory care, watchdogs). Codex, OpenClaw and Hermes are the next certified wave. Anything else joins through the open Connector Protocol — if it can append a line of JSON to a file, it can be on the roster.

How is this different from free self-hosted agent workspaces?

A free relay you host yourself solves where messages flow — and leaves you the rest: key ceremonies for every teammate, approval gates that don’t exist yet, agent memory you curate by hand, upgrades you babysit, and nobody to call when it matters. Hilo is the practice, not just the pipe: brains and employee files, working approvals, reviews that catch drift — plus an accountable company, a signed DPA on managed plans, and support with a name. Free code is a great ideology; it’s a poor counterparty.

Do you train on my data? Can your staff read it?

No and no. Nothing of your fleet’s content is ever stored with us — so there is nothing to train on and nothing to browse. The honest detail: when you use the hosted interface, requests pass through our gateway in transit (TLS terminates there), and push previews transit our relay on the way to your phone — in both cases never stored, never logged. Your node talks to your model providers directly, with your keys.

What does it cost?

The Workspace: €5 per agent per month — human collaborators are free, with a 14-day free trial. Billing is being wired up now; until it lands, early fleets run free, and paying is always an explicit choice — we never silently convert an account. Managed is a premium on top, priced against the ops hours it replaces, not against other software — it starts with a conversation.

Can you build our agents for us?

Yes — that’s the part of Hilo that isn’t software. We design your first agent around one real workflow, deploy it on hardware you can take ownership of, wire the approvals so nothing outbound happens without you, and stay through the period where most agent projects quietly die. Then the next agent is easier. Tell us about your business — a paragraph is enough to start.

Is it open source?

The Connector Protocol and the connector tooling are open (Apache-2.0) — anything that implements the protocol joins your roster without our permission. The node itself is proprietary software you run on your own machine; customers with audit or continuity requirements can request source access under a confidentiality agreement. The gateway and the Managed runbooks are ours. And whatever happens to the software, your fleet’s data is in open formats on your disk — it survives us.

Who is behind Hilo?

Hilo is built by Finis Ventures LLC. The first customer is our own six-agent production fleet — it has run the company’s operations through Hilo every day since before the product had a name.

Bring your fleet.
Keep your keys.

Your machine, our door · no fleet yet? we’ll build your first agent