- a terminal on macOS or Linux, and Homebrew on macOS;
- Node.js or Python 3.10 or later, for the SDK step;
- a git repository on your machine and Claude Code, for the CLI and sessions steps.
1. Sign in and create your project
Open app.to11.ai and sign in with GitHub, Google, or Microsoft. The first time you sign in, onboarding asks for your name and then creates three things: an organization for your company, a workspace inside it, and a project inside the workspace. Name the project after the git repository you will use in step 5, for exampleweb for a repository in a folder called web. Core concepts explains how the three levels fit together.
When onboarding finishes, the dashboard opens on your project’s Sessions page. It is empty for now; you fill it in step 5.
If a teammate invited you, sign in with the email address the invitation went to. You join their organization instead of creating one, and can choose one of its projects, or ask them to create one for you. See Team.
2. Create a chat prompt
In the sidebar, under Build, open Chat Prompts and choose New prompt.-
Set Display name to
Welcome message. The Slug fills in aswelcome-message. -
In the content step, write this system message:
Typing
{{starts a variable reference.nameis a variable your code fills in. - Save the prompt.
latest label points at that version. Your code renders latest.
3. Create an API key
Your code authenticates to to11 with an API key.- Open Settings at the bottom of the sidebar, then API keys under Organization.
- Choose Create API key.
- Name the key
quickstart. Under Where does this key live?, choose Project, then your workspace and project. - On Permissions, keep the Read preset. It lets the key read prompts and skills, which is all rendering needs. Keep the default expiry.
- Review the key and create it.
- Copy the secret. It is shown only once.
proj_. Set both values in your terminal:
4. Render the prompt from code
Install the SDK and render the prompt you created:Ada in place of {{ name }}. Node prints it like this, and Python prints the same list with its own quoting:
format: "openai" shapes the messages for the OpenAI client, so you can pass them to openai.chat.completions.create() with your own OpenAI key. Edit the prompt in the dashboard and publish a new version, run the script again, and the output changes with no change to your code.
5. Link a repository and see a session
The to11 CLI installs your project’s skills into the directories your coding agent reads, and uploads your Claude Code sessions to the project. Install it:0.36.0 or later.
Go to your git repository and run to11:
to11 links the repository to a project, asking one question at a time:
- At
Press Enter to continue in your browser, or Ctrl-C to cancel., press Enter and sign in in the browser. - Because you named the project after the repository, it asks
Found project <workspace>/<project>. Receive its skills here?. Press Enter for yes. If it asksWhich project?instead, choose the project you created in step 1. - Press Enter to accept the skills and the coding agents it installs them for.
- Answer
ytoInstall the to11 plugin for Claude Code?. The plugin keeps your skills current while Claude Code is open. - At
Sync this repository's Claude Code sessions to to11 automatically?, press Enter to keep it off. You sync by hand below.
.to11/skills.yaml, which your team can commit, and .to11/sessions.yaml, which names your project as the place your sessions go and stays out of git.
Now start Claude Code in the repository, ask it one question, and exit:
What you did
You created a project, rendered a chat prompt from code with a project-scoped API key, linked a repository to the project with the CLI, and uploaded a Claude Code session to it.Next steps
Skills quickstart
Write a skill and store it, so everyone who follows the project receives it.
Rendering prompts
Variables, tools, model settings, and sending the result to a model.
Syncing sessions
Send sessions automatically after each response.
API keys
Scopes, rotation, and revocation.