One command, then just write code.
Wilco gives you one coding agent, wilco, and Claude Code, Codex and Grok Build, all on one weekly budget. No key to copy, no config to hand-edit.
What Wilco is
A seat, not an API key.
You buy a seat. The seat carries one weekly allowance, shown as an estimated number of prompts. Claude Code, Codex and Grok all draw on it. You are not paying per token. You are paying for a week's worth of work.
Your tools talk to a small daemon on your own machine, wilcod, which forwards each request to Wilco's gateway. When you use Claude Code, Codex or Grok Build, the client and its agent loop are the official ones, unmodified.
How it works
Where a request goes after you hit enter.
wilco or Claude Code→
local wilcod→
Wilco gateway→
Anthropic, OpenAI or xAI
wilco is Wilco's own coding agent: one terminal agent that reaches every runway on your seat and switches model mid-conversation with /model. It is built from the open-source pi agent (MIT) with Wilco's provider, theme and defaults compiled in. The official CLIs are still here. wilco-claude is a launcher, not a fork: it makes sure wilcod is running, points a private config directory at it, and then execs the official claude binary. Your plain claude keeps whatever account it had. The two do not mix.
Authorization happens once, in the browser, the same way a normal sign-in does. Afterwards your config files hold only a loopback token, scoped to this machine and worthless anywhere else. The seat credential arrives during that authorization. On macOS it is kept in the Keychain and on Windows it is protected with DPAPI. On Linux, where there is usually no keychain to use, it is kept in a file under ~/.wilco that only your user can read.
What gets recorded
The dashboard shows when you made calls, which models they went to, and how much of your weekly allowance is used. Technical logs are kept for billing, abuse prevention and debugging, under access control. Wilco never uses your code or conversations for model training. Requests are processed by the model providers under their own data policies. See the terms and privacy notice.
Runways & models
A runway is a model provider. The CLI is the tool you type into.
Two different things share your seat. Providers are where the models live: Anthropic, OpenAI and xAI today. CLIs are what you run in the terminal: wilco, Wilco's own agent that reaches every runway, plus Claude Code, Codex and Grok Build, all official binaries. Codex is OpenAI's coding agent, not a model; gpt-5-codex is one of the GPT models it drives, and on Wilco it is served by GPT-5.6 Sol and counts as Sol. Any tool that speaks the OpenAI-compatible API can also point at wilcod directly.
| Provider | Models on your seat | Reach it with |
|---|---|---|
| Anthropic · Claude | claude-fable-5-1, claude-fable-5, claude-opus-5, claude-opus-5-5, claude-sonnet-5, claude-haiku-4-5, claude-opus-4-8, claude-opus-4-7, claude-opus-4-6, claude-sonnet-4-6, claude-opus-4-5, claude-sonnet-4-5. Any other claude-* id is passed through as well. | wilco, whose /model offers Opus 5.5, Sonnet 5, Fable 5.1 and Haiku 4.5. Claude Code, wilco-claude, can use any model listed |
| OpenAI · GPT | gpt-6-astra, gpt-6, gpt-5.6-sol, gpt-5.6-luna, gpt-5.6-terra, gpt-5.6, gpt-5.5, gpt-5.4, gpt-5.4-mini, gpt-5.3-codex-spark, gpt-5.2, gpt-5-codex, codex-auto-review, o1 / o3 / o4, gpt-image-2.5, gpt-image-2, gpt-image-1.5, gpt-image-1. wilco-codex asks for gpt-5-codex by default, which is served by GPT-5.6 Sol. | wilco, whose /model offers GPT-6 Astra and GPT-5.6 Sol, Terra and Luna. Codex, wilco-codex, or any OpenAI-compatible client pointed at wilcod can use any model listed |
| xAI · Grok | grok-4.7, grok-4.6, grok-4.5, grok-4.20-reasoning, grok-4.20-non-reasoning, grok-4.20-multi-agent, grok-3-mini, grok-build, grok-composer-2.5-fast, grok-imagine-image-2.0, grok-imagine-video-1.5, plus the -latest aliases and the xai/ and x-ai/ prefixed forms. | wilco, whose /model offers Grok 4.7. Grok Build, grok |
| GLM, Kimi, DeepSeek, Qwen planned | Through the OpenAI-compatible endpoint | Not live yet. They will draw from the same pool when they land |
All runways draw on one weekly allowance. There is no per-provider allowance to buy. The authoritative list is whatever /v1/models returns on your seat; this table is a snapshot and is updated when the gateway changes. The Wilco agent's /model keeps to a curated set; reach the others from Claude Code, Codex, Grok Build or your own client.
Quickstart
About two minutes, most of it waiting on npm.
The install command installs Claude Code for you if it is missing. It connects Codex and Grok Build only when they are already on the machine, so if you want them, install them first with their vendors' instructions. For Codex:
Installed one later? Run the Wilco install command again and it connects it.
Pick a plan on the pricing page and pay by card. Activation is immediate. Check that the dashboard shows your plan as active.
From the Install tab, or copy it here. On macOS or Linux, in a terminal:
On Windows, in PowerShell (see On Windows):
A browser tab opens. Check the email shown on that page is the account your plan is on, then approve this machine. On a server without a browser, open the link on your laptop or phone instead.
Do not want Claude Code on this machine? Add -s -- --skip-claude to the command: wilco works without it, only wilco-claude needs it. On Alpine and other musl-based Linux the Wilco agent does not run (it needs glibc), so those machines get wilco-claude only.
The Wilco agent comes up on your seat. Type /model to switch between Claude, GPT and Grok. That is the whole setup. There was never a key for you to copy.
On Windows
Wilco's command-line tools run on Windows 10 and 11 (x64). Windows 11 on ARM runs them through x64 emulation. Windows PowerShell 5.1 and PowerShell 7 both work. Run the install command in a normal PowerShell window, not one opened with Run as administrator.
The steps are the same as on macOS and Linux. You approve the machine in your browser, the installer downloads and verifies Wilco, then sets up the local proxy, wilcod. It runs in the background and starts by itself when you sign in to Windows. When the installer is done, open a new terminal window: only new windows see the updated PATH.
The commands are the same too: wilco for the Wilco agent and wilco-claude for Claude Code. If Codex or Grok is already installed, you also get wilco-codex and wilco-grok.
Git for Windows is optional. With it, the Wilco agent runs commands in Git Bash; without it, the agent uses PowerShell. Claude Code also works better with Git for Windows installed; Anthropic's Claude Code documentation has the details.
Uninstalling and the other maintenance switches are under The commands. A few Windows specifics are in the FAQ.
The commands
One agent, three vendor CLIs, a few maintenance switches.
wilco is Wilco's own agent. Run it in any project: it reads and edits files, runs commands and switches runway with /model; wilco -p "..." runs one prompt and exits. wilco-claude wraps the official Claude Code: arguments pass straight through, so wilco-claude --help, wilco-claude -p "..." and everything else behave exactly as the official CLI does. The installer prints all three launch commands when it finishes.
| Command | What it does |
|---|---|
wilco | Starts the Wilco agent against your seat: Claude, GPT and Grok models in one session. /model switches runway, /wilco shows the local gateway. |
wilco-claude | Starts Claude Code against your seat. |
wilco-codex | Starts Codex against your seat. Plain codex is untouched. It keeps whatever ChatGPT login you already had, because this runs codex --profile wilco against its own overlay config. |
grok | Grok Build already goes through your seat once installed. wilco-grok is the wrapper form. It also makes sure the daemon is up and steps around a corporate proxy. |
wilco --check | Quick check on macOS and Linux: the local proxy, your plan with its refill or expiry time, the connection to Wilco, and the commands this machine really has. It starts the local proxy if it is not running and changes nothing else. It needs neither Claude Code nor any of your weekly pool. Start here when something looks wrong. On Windows, use the installer's -Check below. |
wilco-claude --check | Claude Code specific check: what the launcher configures, where your credential is kept (keychain / DPAPI / file), whether the daemon sees a proxy, what leaked into your shell rc, and which gateway a plain claude would hit. |
install.sh --purge | Removes keys and env vars that earlier installs left in your shell rc, so plain claude stops picking up our gateway by accident. Wilco keeps working. |
wilco-claude --route main | Shows or sets the gateway route. There is one route today. |
install.sh --uninstall | Puts the machine back: rolls back the Codex and Grok config files (only our marked block; your own settings are kept verbatim), removes the overlay, the wrapper commands, the config directory and the shell rc entries, stops the daemon and clears the keystore entry. |
The two install.sh switches go on the end of the install command:
curl -fsSL https://wilco.run/install.sh | bash -s -- --uninstall
On Windows, the installer takes them as PowerShell switches, plus -Check, a read-only check. -Purge looks for leftover keys in your user environment variables and PowerShell profile instead of a shell rc, and -Uninstall also removes the autostart entry and the PATH entry:
& ([scriptblock]::Create((irm https://wilco.run/install.ps1))) -Purge
& ([scriptblock]::Create((irm https://wilco.run/install.ps1))) -Uninstall
wilco-claude says so when it starts.Commands & paths
Where things land on disk.
| Path | What it is |
|---|---|
~/.wilco/ | Config directory: the loopback token and the isolated client environment. On macOS and Windows the seat credential is not here; it lives in the OS keystore. On Linux it is kept here, in a file only your user can read. |
~/.wilco/wilcod.log | Daemon log. Every way the daemon gets started writes here, so this is the first file to look at. |
~/.local/bin/wilco | The launchers: wilco, wilco-claude, wilco-codex, wilco-grok. /usr/local/bin is used instead on some systems. |
~/.wilco/agent-bin/ | The Wilco agent binary and its assets. |
~/.wilco/agent/ | Wilco agent settings, sessions, extensions and themes. |
| shell rc | A marked block, # >>> wilco path >>>, that puts the launchers on your PATH. It goes into the startup files you already have, plus ~/.zshrc or ~/.bashrc for your own shell, and ~/.profile only if you have no login file at all. ~/.bash_profile is never created. fish gets ~/.config/fish/conf.d/wilco.fish. --uninstall removes exactly that block and deletes any startup file Wilco created that is then empty. |
No sudo is needed for a normal install. Apart from the launchers, which go to /usr/local/bin only when your user can already write there, everything stays in your home directory.
On Windows, Wilco's files live under %USERPROFILE%\.wilco\. The launchers are in its bin folder, which the installer adds to your user PATH, and an entry named Wilco in your startup apps starts the local proxy when you sign in. No administrator rights are needed.
Skills & plugins
They work, because this is the real Claude Code.
Because wilco-claude execs the official binary, skills, plugins, MCP servers and CLAUDE.md files all behave normally. The one thing to know is where they are read from: Wilco runs with its own config directory, so per-user files you had under the default Claude Code home are not picked up automatically.
Project-level files such as CLAUDE.md and .claude/ in your repo are unaffected and work as usual.
~/.wilco/ in the same layout Claude Code expects, then restart wilco-claude.The weekly pool
One allowance, every model.
Your plan carries a weekly allowance, shown as an estimated number of prompts. One prompt is one instruction plus the agent work it sets off. The counts on the pricing page are for Claude Opus 5.5 at the median prompt size we see on real Claude Code sessions: Sonnet goes about twice as far and Fable about a third, while long unattended runs and very large contexts use more. Spend it however the work needs.
The pool refills seven days after the window opened, not on a calendar week. The dashboard shows the exact reset time. Unused budget does not roll over. A plan can use at most one weekly pool for every 7 days, about 4.3 pools in 30 days, which is the monthly figure on the pricing page; see Limits.
Every runway shares it
One seat, one pool, every provider. Claude, GPT and Grok models all draw on the same weekly allowance, whichever CLI you use. There is no separate allowance per provider and no per-engine surcharge.
Limits
What can pause your requests, and when they come back.
A plan has three limits. There is no shorter window and no separate cap for any one model.
| Limit | How it behaves |
|---|---|
| Weekly pool | Refills seven days after the window opened. Running it out pauses requests until then. Every model draws on it. |
| 30-day cap | One weekly pool for every 7 days (about 4.3 per 30 days), the monthly figure on the pricing page. You only meet it if you have used nearly every earlier week in full; the dashboard shows it when it is close, and requests then resume when the 30-day window rolls over. |
| Plan term | 30 days, with no automatic renewal. When it ends, requests stop and nothing is deleted; renew and the same machines work again. |
The weekly pool and its exact reset time are under Billing.
Billing & payment
By card, in US dollars.
Paying
Plans are priced in USD and paid by card; your card statement shows ZEYLANE* WILCO. A plan runs 30 days and does not renew on its own. Activation happens as soon as payment lands. There is no manual review step.
Upgrading
Upgrading is a difference, not a second purchase: you pay the new plan's price minus what your current plan still has left. The new plan starts that day with a full term and a full weekly budget.
What is left is the value of your remaining days, with one rule for the current week: it counts only the part of this week's budget you have not used yet. If what is left is worth more than the new plan, you pay nothing and the extra becomes additional days on the new plan. Moving to a smaller plan is not self-service.
Your access key does not change when you upgrade, so no machine needs reinstalling.
Credit and rebate
Two things can sit in your wallet:
- Credit balance: spent before a card is ever charged, on renewals and upgrades alike. It can also be converted 1:1 into extra days on your current plan, whole days only.
- Referral rebate: a share of what people you invited pay. Each share is held for 7 days, then lands in "ready to move"; move it to your balance to spend it.
Both live under Billing, and your invite link is under Referrals.
Promo codes, gift codes and invites
- Promo codes: enter one under Have a promo code? when you pay, or open a link that carries one. First-plan codes take up to 10% off your first plan; renewal codes take up to 5% off renewing a plan you have had before. Every code has an end date and a limited number of uses, and each account can use a code once.
- Gift codes: redeem one under Billing → Gift code. It adds the days on the code to its plan straight away, with no card and no checkout: they are added to that plan if you have it, or the plan starts today if you have none. If a different plan is active, contact support. Each account can redeem a code once. Gift days have no cash value, so they are not credited when you upgrade.
- Invites: someone who signs up through your invite link gets a discount on their first plan (currently 10%), and you earn a share of what they pay from then on.
- Invite bonus: on top of that share, you get a one-time bonus of 5% of the first payment of each person you invite. It is added to your balance once that payment has stood for 7 days with no refund, and each person counts once.
- One discount per order: when a promo code and an invite both apply, you get the larger one.
- A discounted order is paid through your wallet: the discounted amount is charged to your card as credit and the plan is activated from it in the same step, so the payment is recorded as a credit top-up.
Downgrading
Moving to a smaller plan is not self-service. Open a ticket and we will sort it out.
Machines
How many computers one seat covers.
There is no fixed machine limit on a seat today. The Settings tab lists every machine currently signed in, with its IP and when it was last seen, and you can remove any of them.
New laptop? Run the install command on the new one and approve it, then remove the old machine there. Removing a machine takes it off the list. It keeps the access key it already has, so if a machine is lost or not yours, remove it and open a ticket to have your key reset; your machines then need to be authorized again.
Password and sessions
Forgot your password? Use Forgot password? on the sign-in page: a one-time link goes to the email on your account and works for 30 minutes. Signed up with Google? There is no password; use Continue with Google. Under Settings you can change your password and use Sign out everywhere. Changing the password signs out every browser session at once; Sign out everywhere stops every session from renewing, but a session that is already open can keep working for up to a day until its sign-in expires. Neither touches your machines: they use the access key, not your password.
Troubleshooting
Start with wilco --check (on Windows, the installer's -Check). It answers most of these.
Install and PATH
| Symptom | Fix |
|---|---|
wilco: command not found | Open a fresh terminal. If it persists, confirm ~/.local/bin or /usr/local/bin is on your PATH. |
Windows: wilco is not recognized | Open a new terminal window: the window you installed from keeps its old PATH. If it persists, check that %USERPROFILE%\.wilco\bin is in your user PATH. |
claude: command not found | Only wilco-claude needs it. The installer tries to set it up; otherwise run curl -fsSL https://claude.ai/install.sh | bash (on Windows, irm https://claude.ai/install.ps1 | iex) and run the Wilco install command again. |
| npm permission errors | Fix your npm global prefix as npm suggests. Do not blanket-sudo the install. |
curl or irm cannot connect | Check network and DNS, and that the URL really is https://wilco.run. |
Authorization
| Symptom | Fix |
|---|---|
| Link expired | Re-run the install command for a fresh one. |
| Device limit reached | Remove a machine under Settings, then authorize again. |
| Wrong email on the approval page | Sign out there and sign back in with the account your plan is on. |
| No active plan | Subscribe first on the pricing page, then authorize. |
401 / 403 / stopped working
| Symptom | What it means |
|---|---|
| 401 unauthorized | The local config is missing or stale. Re-run the install command and approve again. |
| 403 forbidden | Plan expired, or the weekly pool is spent. The dashboard says which. |
| "Cannot continue" mid-task | Usually the weekly pool, or the 30-day cap after about four full weeks (see Limits). Check Billing. Neither is an outage. |
Plain claude hits our gateway | An old install leaked env into your shell rc. Run install.sh --purge (see The commands). |
Behind a corporate proxy
Works out of the box. Whatever HTTP_PROXY / HTTPS_PROXY / NO_PROXY your shell had at install time is written into the daemon's startup config, and each launch command adds 127.0.0.1 and localhost to NO_PROXY before starting, so your own traffic to the daemon never goes out through the proxy. wilco-claude --check shows whether the daemon picked up a proxy. On Windows, see the FAQ if your proxy is set only in Windows settings.
Model list looks wrong
Run wilco-claude --check and confirm the gateway is the local address (127.0.0.1). If it is, upgrade Claude Code itself. An old client can mis-render newer model names.
FAQ
Is this the real Claude Code?
Yes. wilco-claude is a launcher around Anthropic's official binary, and we do not modify it. On some cross-provider routes, such as Claude Code using a GPT model, the gateway converts the request between formats and adds short compatibility instructions. wilco, the Wilco agent, is the one thing we build ourselves, on top of the open-source pi agent.
Can I use my own Claude account alongside it?
Yes. That is the point of the separate config directory. Plain claude keeps its own login; wilco-claude uses your seat. If they ever get tangled, curl -fsSL https://wilco.run/install.sh | bash -s -- --purge separates them again. On Windows, the installer's -Purge switch does the same (see The commands).
Does it work on Windows?
Yes, on Windows 10 and 11 (x64), in PowerShell. On Windows has the install command and what it sets up. A few things to know:
- Wilco's Windows programs are not code-signed yet, so Windows or your antivirus may show a security prompt.
- The installer does not run from an administrator window. It says so and stops; open PowerShell normally and run the command again.
- If your network needs a proxy that is set only in Windows settings, the local proxy cannot connect through it. Set
HTTPS_PROXYin the PowerShell window before you run the install command.
VS Code and the Claude desktop app are still coming.
What about Codex and Grok?
Both work on your seat. The install command connects Codex and Grok Build when they are already installed; if you add one later, run the install command again (see Quickstart). They share your weekly pool with no surcharge. See Runways & models for which provider each one talks to. VS Code and the Claude desktop app are coming.
Can I point my own tool at it?
Yes. Anything that takes a base URL (Cline, Roo, a script you wrote) can use the local address with the loopback token. You are never issued a key that works off this machine, so there is nothing to leak into a repo.
How do I remove it?
Run curl -fsSL https://wilco.run/install.sh | bash -s -- --uninstall. It rolls the machine back to how it was, including the Codex and Grok config files it touched.
On Windows, run this in PowerShell. It does the same and also removes the autostart entry and the PATH entry:
Do you see my code?
Requests pass through our gateway, so in transit, yes. That is what a gateway is. We keep metadata and technical logs, not your source, and we never use your code or prompts for model training. The model providers process requests under their own data policies. The privacy notice is specific about this.
What happens when my plan expires?
Requests stop; nothing is deleted. Renew and the same key on the same machines starts working again.
Refunds?
If you ask within 7 days of your first purchase and have used less than 20% of your first week's pool, we refund it in full (once per account). Otherwise we refund the unused days minus the allowance already used, and if Wilco is down for more than 24 hours in a row on our side, you choose extra days or a pro-rata refund. The terms of service have the details.
Support
Tickets first, email if you prefer. Real people either way.
Open a ticket under Support in the cockpit. Pick what it is about (billing, technical, account and machines, or anything else), describe it, and attach screenshots or logs if you have them. Your account email and plan travel with the ticket, so there is nothing to quote. Replies land in the same thread and, if you like, in your inbox. We usually reply within 24 hours.
Quick question, or not signed in yet? Use the help button in the corner of every page. It answers the common questions on the spot, and Chat with us opens a live chat with the same people. Replies show up in that chat window, on any page of this site in the same browser. If you might close the page before we answer, put your email in the chat so we can write back. Anything that needs a paper trail, like a charge or a key reset, is better as a ticket.
Prefer email? Write to support@wilco.run from the address your plan is on. For a charge or a machine, add the technical details from the bottom of Settings.
For anything that looks like a billing error, give the exact amount and the time. That is usually enough to find it in one pass. A closed ticket can be reopened from the cockpit for 30 days.