Pilos Agents runs the agent on your machine and turns your best conversations into scheduled workflows. Pilos Rooms puts your whole team inside a single live session — watching, steering and approving together.
Free. Open core, MIT licensed. Agents runs entirely on your machine.
A native app on top of Claude Code. Build a team of agents, give them a real terminal and 22 tool integrations, then turn any conversation that worked into a workflow that runs on a schedule.
One agent, many people. The runner stays on a developer's machine; everyone else joins a room in the browser to follow the work live, queue messages, approve the risky commands and take the wheel when it is their turn.
Build a workflow once from any AI conversation. Run it on a schedule. Never prompt the same thing twice.
A drag-and-drop canvas for AI pipelines. Connect agents, tools, conditions and loops — 13 node types, auto-layout, undo and redo. No code.
Run any workflow every 15 minutes, hourly, daily or weekly. Nightly code reviews and morning reports happen while you sleep.
Assign a PM, an Architect, a Developer or a Designer to each step. They disagree with each other, and the output is better for it.
Conversations and runs live in SQLite on your disk. No cloud sync, no telemetry, nothing leaving the machine unless you ask for it.
GitHub, Supabase, Slack, Jira, Figma, Kubernetes, Terraform and 15 more MCP templates. Your agents read, write and act on real data.
A real shell in the window. Agents run commands, read the output and react — no tab-switching between your terminal and your AI.
Chat with your agents about any task — a report, a code review, a research brief. Use it the way you already do.
Turn a conversation that worked into a reusable graph. Add parameters, conditions and connected tools, visually.
Every 15 minutes, daily at 9am, or weekly on Monday. Pilos runs it and delivers the result where you asked.
Every run is in your history. Improve the workflow as you go, share it with the team or keep it to yourself.
Everyone uses AI to write the code. Turning that work into a standup, a report or a timesheet is still copy-paste. Pilos reads your commits — including what you have not committed yet — and writes it for you.
Daily Standup, Detailed Work Report, Manager Summary or an invoice-ready Timesheet — all from the same git activity. Runs on Pilos Cloud with no setup, or entirely locally with your own key.
The agent runs on one machine. Everyone else opens a link. They see every message, every diff and every command as it happens — and when the agent wants to do something irreversible, it stops and asks the room.
Claude — Migrated the refund handler to the new provider and updated the tests. Ready to push the branch.
$ git push origin refund-flow --force
git-push
Pushing rewrites history on a shared branch.
A push, an rm -rf, a deploy, a publish, a write outside the
repo — the agent stops and the room decides. Who may approve is set per rule, and the
decision is recorded with the name of whoever made it.
Exactly one person holds the wheel. Everyone else can still type — their messages are queued and visible to the room until the driver releases them. Handing over writes a summary of where things stand.
A room is an append-only log. Someone joining an hour late scrolls back through exactly what the first person saw — the same events, in the same order, not a summary of them.
The runner connects outbound only. Nothing opens a port on your machine, no one gets a shell, and your environment variables never leave it. The room sees the events the runner sends and nothing else.
Read the Rooms docs| Role | Can do |
|---|---|
| Owner | Everything: invite and remove people, change roles, take the wheel at any time, approve anything the policy allows an owner to approve. |
| Driver | Holds the control lock. Sends messages straight to the agent, releases other people's queued messages, pauses and resumes the run. |
| Reviewer | Approves and rejects the calls the policy assigns to reviewers. Can ask for the wheel. Messages are queued until released. |
| Viewer | Sees everything, including every approval card and every decision. Sends nothing. |
Driving is a position, not a replacement for a role: a reviewer who takes the wheel still approves what reviewers approve.
The Agents core is MIT licensed and fully auditable. No black boxes, no telemetry, no vendor lock-in. Run it, fork it, change it.
Source-available under BUSL-1.1. Both halves ship in every download, at no cost.
No plans, no licence key, no card. Every feature is unlocked. An account is optional — the app works the moment it finishes downloading.
Sign up to sync your work across devices, pair a phone and use Pilos Cloud for the Work Day Reporter.
Already have an account? Sign in inside the app under Settings → Account.
Powered by your own Claude Code install — no extra API key needed.
Agents is a desktop app you install: it runs the agent locally and is built around one person automating their own work. Rooms is a web app: the agent still runs on one person's machine, but the session is shared with everyone you invite, with approvals and a control lock so several people can work with it without talking over each other. They are separate products and you can use either on its own.
Yes, for both. Install it with npm install -g @anthropic-ai/claude-code
and authenticate with your Anthropic account. No separate API key is needed — the CLI
handles it. Claude Code is the only agent either product drives today.
Agents is free in full — no licence key, no trial clock, no card, and no paid tier to upgrade to. Rooms is free while it is in open preview. If we ever introduce pricing we will say so clearly and give plenty of notice rather than switching a paywall on overnight.
Everything the runner sends: the agent's messages, its plan, the diffs, the commands it runs and their output, and every approval with the name of whoever decided it. They cannot see anything else on the machine. The runner connects outbound only, opens no port, gives nobody a shell, and never transmits environment variables or secrets.
For Agents, on your disk: conversations and runs are stored in local SQLite, and nothing is synced unless you create an account and ask for it. Rooms is different by design: a room's event log is stored on our server so that someone joining late can replay it, and that log includes the diffs the agent produced and the output of the commands it ran. Credentials are stripped before anything is sent, but the code in a diff is not — see the privacy policy for exactly what is kept.
The Agents core is MIT licensed, so yes — fork it and build it yourself. Rooms is a hosted preview today and is not packaged for self-hosting yet.
Agents is free to download and runs on your machine. Rooms is free while it is in preview.
macOS, Windows and Linux · requires Claude Code · View on GitHub