Skip to main content
The goal of this quickstart is a verified local setup and a trustworthy skill inventory. SelfTune should tell you what it knows and identify where it does not have enough evidence yet.
You need Node.js with npx or Bun, an installed AI agent, and at least one agent skill. SelfTune uses your existing agent environment for interactive reasoning; local setup does not require a separate model API key.
Copy the setup prompt from Set up with your agent. Your agent will inspect the environment, choose the right installation path, and stop when it needs permission.

​
Manual setup

1

Install the agent skill

Restart or reload your agent if it does not discover newly installed skills during the current session.
2

Initialize SelfTune

Ask your agent:
The agent should detect supported harnesses and explain the changes it intends to make.
3

Check the runtime

Resolve reported errors before continuing. Warnings about missing history or no observations can be valid on a new installation.
4

Import existing local evidence

Sync reads supported local agent histories and normalizes skill evidence into SelfTune’s local data store. It does not make a remote upload part of this quickstart.
5

Inspect the inventory

A useful result distinguishes among observed health, a skill that needs attention, and a skill with insufficient evidence.

​
Verify the result

You are done when all of these are true:
  • doctor reports no blocking runtime or permission error;
  • SelfTune finds the skills you expect it to find;
  • status returns a health or no-data state for each discovered skill; and
  • no remote upload, publication, credential, or background-service change happened without your approval.
A no-data result is valid on a fresh installation. Use the skill naturally, then run sync and status again after the agent session ends.

​
Open the local dashboard

The dashboard uses the same local data as the CLI. It gives you a visual skill inventory, health details, and links to the next appropriate action.

​
If setup does not work

Run doctor again and keep its exact output. Then follow Troubleshooting for runtime, discovery, permissions, and no-data problems.