For heavy users of Claude Code and Codex
Everything you capture becomes work. One board for you and your agents.
Mail, meetings and saved links turn into tasks. Delegate them to Claude Code or Codex, take them back when you want, and get called only when a decision is yours.
- Free and open source
- Runs on your machine
- Uses the plans you already pay for. No API key
Mobile appcoming soon
Open source
Every line is readable. The repository opens at launch.
Self-hosted
Your Mac or a small VPS. Your files, your plans. Nothing phones home.
Built by a heavy user
Made by someone who runs Claude Code and Codex all day, to stop babysitting them.
Mail, meetings and links turn into tasks.
The inbox desk sorts what came in and drafts the reply. Meeting next steps become tasks. A link you share is read, summarized and filed.
mail · team chat · meetings · WhatsApp are connectors
You and the agents, same board.
Your tasks have a check. Theirs have a conversation. Delegate yours, take one back, accept a branch when it’s good, and watch the plan you pay for while it happens.
Also on the board: memory you approve, links you saved, your servers and what the APIs cost. Same screen.
Everything that runs, by project.
A task in the queue, the Claude Code session open on your Mac, Codex in a tmux on your VPS: one list per project, with where each one runs, what it is doing and what it costs the machine. Nothing shows up twice.
They ask before they push. They stay in their lane.
When a decision is yours, you get one message with one question, and you answer it from the chat you already have open. The yes goes back to the task. Nothing ships without it.
A cap per plan
Three on Claude Max, two on Codex. The rest waits in line.
The right lane
Work never runs on your personal plan. The folder picks the account.
Read-only first
Agents look before they touch. Say “code” and they get a branch.
Ask before pushing
Push, merge and deploy wait for your yes. One question per message.
Memory with consent
The librarian proposes; you approve. Secrets never reach the index.
Logins stay home
Password and 2FA on the door. Your subscriptions never leave the machine.
Get it the day the repo opens.
One email at launch, with the repository link. No newsletter.
- What does it cost?
- Nothing. Self-hosted Wilco is free and open source. You keep paying only for the Claude and ChatGPT plans you already have.
- Is my code safe?
- It runs on your machine. Files on disk, one login with 2FA, no telemetry. Agents can't push, merge or deploy without your yes.
- What about my subscriptions and logins?
- Wilco drives the Claude Code and Codex you're already logged in to. No API key, nothing proxied, and your logins never leave the box.
- Why not Orca or Happy?
- They start at the terminal, with the task already written. Wilco starts where work shows up (mail, a meeting, a link) and calls you only when a decision is yours.
Step away. Stay on the radio.
On your phone, the same board and the same conversations, with a push when an agent needs you. On your Mac, Wilco lives in the menu bar: it drafts the reply into the field you’re typing in, translates what you selected and summarizes a link before you open it.
Mobile appcoming soon
⌃⌥Space in any text field: what they want, three drafts, paste with one key. Nothing is sent until you send it. soon
Yours, on your machine.
Wilco sits next to the Claude Code and Codex you already have installed. It doesn't ask for API keys and it doesn't phone home.
data/
agents.db · queue, runs, log
usage.json · quota per plan, live
breakdown.json · where it went
memory-index.db
saved.db
web/ · the dashboard
worker/ · runs agents on this machine
collector/ · reads usage, servers, spend
$ wilco status
4 running · 3 queued · Claude Max 62% (resets 1h12)
0
API keys. Agents run on the Claude and ChatGPT plans you already pay for.
100%
of your data stays on your machine. JSON and SQLite in one folder.
3
agents at once on a Claude plan, 2 on Codex. The rest waits in line.
1
command to start.soon
01
Run it on your box
A Mac that’s usually on, or a small VPS. Files on disk, a SQLite queue, one login with 2FA.
02
Connect your plans
Log in to Claude Code and Codex on that machine, the way you always do. Nothing is proxied through anyone else’s server.
03
Call it in
From the web, the CLI or your phone. Pick the account and the role. Agents pick up work as slots free up.
Straight answers.
Is it free?
Self-hosted Wilco is free and open source. A hosted version may come later for people who don't want to run a server. The core stays free.
Which agents does it run?
Claude Code and Codex today, under your own subscriptions. Each account gets a cap; the queue respects it.
Does it need my API key?
No. Wilco runs the CLIs you already log in to and reads the usage those plans report. If you also spend API credits elsewhere, the Spend page can track those.
Can an agent push to main or deploy?
Not without you. Agents work on their own branch in read-only mode unless you say “code”, and push, merge and deploy wait for your yes. You give it in the chat, one decision per message.
What about work and personal accounts?
They never cross. Each context has its own accounts and memory scope; a work task won't run on your personal plan, and Delegate picks the account from the folder the task lives in.
Where does my data live?
In a folder on the machine you run Wilco on. JSON files and SQLite. No external database, no telemetry. Secrets are stripped before anything is indexed.
Can I use it from my phone?
Yes. Wilco on your phone shows the board and the quota, pushes when an agent needs you, and takes your reply. It talks to your own instance.
How is this different from Orca or Happy?
They start at the terminal, with the task already written. Wilco starts where the work shows up (mail, a meeting, a link) and ends with you only when a decision is yours. Plus the rules the agents follow: caps, lanes, consent.
Why “wilco”?
Radio shorthand for “will comply”. You call it in; the agents say wilco.