PiLot Studio needs two pieces: the PI coding agent (the runtime) and the PiLot Studio extension (the interface). Install PI first, then the extension, then connect a provider.

Before you install

VS Code 1.105 or newer. PiLot Studio 3.0 uses PI 1.4, which needs Node 22.19. VS Code 1.105 is the first release that bundles that Node version.
Node.js 22.19 or newer and npm. Needed to install PI if you do not already have it.
A model provider. An API key for Anthropic, OpenAI, Google, OpenRouter, or another supported provider; an OAuth account; or a local OpenAI-compatible server such as Ollama.

1. Install the PI CLI

In a terminal, run:

npm install -g --ignore-scripts @earendil-works/pi-coding-agent

Then confirm that VS Code's integrated terminal can find it:

pi --version

The version must be 1.4.0 or newer. If pi is installed somewhere unusual, note the absolute path; you can enter it in pi-agent.binaryPath later.

If you already use Pi through the CLI, you can keep your existing ~/.pi/agent directory, credentials, packages, and sessions. PiLot Studio reads the same files.

2. Install PiLot Studio

Install the extension from one of these sources:

After installation, reload the window if VS Code asks. Open a folder rather than a single file; PiLot Studio uses the workspace root for project context, @ file mentions, and project-local .pi configuration.

Click the PiLot Studio icon in the Activity Bar, or press:

Windows/Linux: Ctrl+Shift+Alt+P
macOS: Cmd+Shift+Alt+P

The panel opens on the Conversation tab. If the extension cannot find pi, it shows an error with the path it tried. See Troubleshooting.

4. Connect a provider and choose a model

Open the Providers tab (the key icon, or press 4).

If your provider supports OAuth, choose Sign in and complete the browser flow.
Otherwise paste an API key and save it. PiLot Studio writes the credential to PI's auth.json and never sends it to Printaga.
For a local or custom endpoint, use Add a custom provider and fill in the name, base URL, and model IDs.

Go back to the chat header and open the model picker (or press 3 for the Models tab). Pick a chat model. Models that support reasoning also expose a thinking-level control in the header.

PiLot Studio does not proxy model requests. Your prompt and the files you attach go from your machine to the endpoint you selected. Review the provider's data policy before attaching sensitive code.

5. Send a first prompt

In the composer, try:

Summarize this workspace and identify its main entry point.

Press Enter to send, or Shift+Enter for a new line. The reply streams into the conversation. Tool calls such as file reads, searches, edits, and shell commands appear as collapsible blocks.

Useful first checks:

Ask a question about a file and confirm the agent can read the workspace.
Select a few lines in an editor, right-click, and choose Explain Code with PI.
Type @ in the composer and confirm workspace files appear.
Open History and confirm the new session is listed.

6. Keep everything updated

PiLot Studio checks for a newer PI release and package updates in the background every six hours. The header shows an update button when something is available.

PiLot Studio: update through the VS Code Extensions view.
PI CLI: run pi update, or click the header update button.
Packages: run pi update --extensions, or use Update in the Packages tab.
Automatic PI updates: set pi-agent.autoUpdate to true if you want PiLot Studio to open a terminal and run pi update when it detects a newer release.

First-run checklist

pi --version prints 1.4.0 or newer inside VS Code's terminal.
The PiLot Studio view opens without a "binary not found" error.
At least one provider shows API key valid or OAuth active.
A model is selected in the chat header.
A test prompt returns a response.
The session appears in History.

Where PiLot Studio stores things

By default, PI uses ~/.pi/agent. PiLot Studio reads and writes the same directory:

File or folder	Contents
`auth.json`	Provider API keys and OAuth credentials.
`models.json`	Custom provider and model definitions.
`settings.json`	PI settings, including codemode options.
`mcp.json`	Global MCP servers and `autoEnableCodemode`.
`trust.json`	Project trust decisions shared with the Pi CLI.
`sessions/`	Session files shown in the History tab.
`voice-models/`	Downloaded local dictation models.

Change the location with pi-agent.agentDir. Reload the VS Code window after changing it. A project can also add .pi/settings.json and .pi/mcp.json; those only load when the workspace is trusted.

Next steps

The PiLot Studio workspace — what every tab and button does.
Chat, context, and editor actions — mentions, attachments, and slash commands.
Providers, models, and thinking — connect more providers and tune models.

Install and first run