Developers
Projects
Connect a code folder to Capy and hand development tasks to Kapi, to be done on a dedicated cloud computer or by the coding agent on your own computer.
A project is a folder on your computer (it can hold one or more git repositories, or ordinary directories) plus its run settings. Once it's connected to Capy, you can tell Kapi "Fix the error on the login page in project XX", and Kapi sends off a development task, checks it when it's done, and hands the changes back to you as a git branch.
Development tasks can run in two places:
| Where it runs | Good for | Where the code goes |
|---|---|---|
| Agent on my computer (default) | You already use a coding agent such as Claude Code or Codex | The code never leaves your computer, and changes stay on a local branch |
| Cloud computer | You want tasks to run in the background without tying up your computer | The folder is uploaded to this project's own cloud computer, separate from the one that does your spreadsheets |
Where to find it
Projects in the side menu, or Settings → Connections and rules → Projects. A project page has these tabs:
- Overview: cloud computer status, hourly price, last sync time, and this project's tasks;
- Run settings: where tasks run, cloud computer size, sleep, thinking effort, and web tools;
- Network: outbound allowlist, secrets, and intranet access;
- Terminal: open a command line on the cloud computer directly.
Local agent projects have only Overview and Run settings.
Add a project
In the desktop app: in the project list, tap Add folder at the top right → choose a folder → choose Where tasks run:
- Agent on my computer: just records the folder; nothing is uploaded;
- Cloud computer: the button changes to Add and upload, and the whole folder is uploaded.
From the command line (see The capy command line):
capy project add [folder] [--name name] [--size 4c8g] # cloud computer project, full upload the first time
capy project add --runner agent # local agent project, no upload
On your phone and on the web you can view projects, change settings, open a terminal, and follow task progress, but adding folders and syncing must be done on a computer.
Sync code (cloud computer projects)
Tap Sync on the project page in the desktop app, or run capy sync in the folder. Only what changed is uploaded; capy run also syncs once automatically before sending off a task.
Before the first upload, you see a list: how many files and commits each repository has, whether there are uncommitted changes, the total size, and what .capyignore excludes. Nothing is uploaded until you confirm.
It stops if it finds files that look like secrets: if a file name looks like a secret (.env, *.pem, id_rsa, and so on), or the content contains something formatted like a private key or cloud service token, Capy stops without uploading a single byte, and asks you to confirm each one as "not a real secret" before it continues.
- Put anything you don't want uploaded in
.capyignoreat the folder's root (same syntax as.gitignore); - Each upload can be at most 200 MB.
Run settings
| Setting | Options |
|---|---|
| Where tasks run | Cloud computer / Agent on my computer |
| Cloud computer size | 1 CPU · 2 GB, 2 CPU · 4 GB, 4 CPU · 8 GB (default), 8 CPU · 16 GB. Changing the size switches to a new cloud computer |
| Sleep | After 5 minutes idle (default) / Never (always billed) |
| Thinking effort | Low, Medium (default), High, Max; you can also let Kapi raise it on its own depending on difficulty |
| Web tools | Off by default. Before you turn them on, you're warned that web search and reading web pages bypass the outbound allowlist and could carry code out |
The larger the size, the higher the hourly usage. The figures shown on the page are what count.
Network and secrets
To keep code from being carried out, outbound network access for cloud computer projects is limited to an allowlist by default:
- Common package manager mirrors in China (npm, PyPI, Go, and others);
- Capy's own file addresses;
- The domains of secrets you've configured;
- Domains you add yourself.
You can also switch it to Unrestricted. Tasks and the terminal won't start until the outbound rules are in effect.
There are two kinds of secrets:
- Header secrets: added to requests automatically at the exit, by domain. Their values can't be seen on the cloud computer or in the terminal;
- Environment variable secrets: given to the agent when a task runs, only in allowlist mode, with the value masked in logs.
Once a secret is saved, its value can't be viewed again; you can only replace it or delete it.
Reach the company intranet
If a task needs to connect to a database or service on your company's intranet, you can run a relay on an always-on computer inside the intranet (get IT's approval first):
capy login
capy relay register --name server-room
capy relay allow db.corp.example:5432 # one host:port per entry, no wildcards
capy relay start # keeps running, only connects out to Capy, reconnects automatically
Then pick that relay in the project's Network → Intranet access.
- Only projects whose outbound network is in allowlist mode can turn on intranet access (the page has a one-tap Switch to allowlist and turn on);
- TCP only; programs on the cloud computer connect directly to intranet addresses;
- Which addresses can be reached can only be changed on the relay computer, and changes take effect immediately;
- Every connection is logged. In Settings → Connections and rules → Relays you can see whether a relay is online, its list, and the projects that use it, and you can revoke it.
Terminal
Use Open terminal on the project page or the Terminal tab (on computer, web, and phone), or run capy shell in the project folder. It opens a command line on the cloud computer in /workspace, where the agent's files and logs live.
- Opening a terminal wakes the cloud computer, which is billed while awake;
- After 5 minutes with no input or output it disconnects automatically; tap Reconnect to resume;
- On the phone there's a row of Esc, Tab, Ctrl, and arrow keys.
Send development tasks and get changes back
- Tell Kapi which project and what to do, or run
capy run "Fix the error on the login page"in the folder. - Use
capy status <task>to see the status and the changes in each repository, andcapy logs <task> -fto follow the run log; you can also check the progress page in the app. - The agent commits on a
capy/<task>branch in each repository. Kapi checks the work against the delivery notes, the changes, and the test results, and sends it back to be redone if it's not up to standard. - Cloud computer projects: run
capy pull <task>in your local folder, and each repository gets a new local branch for you to review and merge. Local agent projects: the changes are already on a branch in your repository, so there's nothing to pull.
Sleep
Sleep on the overview page puts the cloud computer to sleep by hand; the command is capy sleep. If a task is running, choose Stop tasks and sleep (capy sleep --cancel).
Tell Kapi what this project is
You can edit How Kapi knows this project directly on the overview page: a project description, what people call it in chat, which file issues are recorded in, and which chats it's linked to. Once that's filled in, when someone in a group says "the admin login is down again", Kapi knows which project they mean.