PiLot Studio is a VS Code extension that puts a visual interface on the Pi coding agent. It gives you a chat sidebar, a session browser, provider and model settings, an MCP manager, package and skill panels, and local voice dictation without leaving the editor.

It is not a hosted service and it does not bundle a model. PiLot Studio loads the PI runtime from your own PI installation, so chat requests go directly from your machine to the provider you configure.

What you get

Chat in the editor. Ask questions, request edits, mention workspace files with @, attach images, and apply code blocks to the active editor.
Sessions you can branch. Browse past conversations, rename them, delete them, export them, or fork a new branch from any user message.
Bring your own provider. Connect Anthropic, OpenAI, Google, OpenRouter, Ollama, or an OpenAI-compatible endpoint, choose a model, and set a thinking level.
MCP without the terminal. Add global or project MCP servers, inspect connection state and tools, choose how each server is exposed, and sign in to OAuth servers.
Pi packages, skills, and tools. Install packages from npm, enable or disable individual skills and packages, choose a tool preset, and inspect the live system prompt.
Local voice dictation. Download a whisper.cpp model once, then transcribe on your device.
Git and diagnostics helpers. Draft a commit message from staged changes and capture a local diagnostics log when something goes wrong.

Requirements

Requirement	Minimum	Notes
VS Code	1.105	The first release whose extension host bundles Node 22.19.
Node.js	22.19	Used by the extension host and required by PI 1.4.
PI CLI	1.4.0	PiLot Studio loads the PI SDK from your installation.
OS	Windows, Linux, macOS	Voice dictation is available on x64 and arm64 builds.

The extension checks the Node version during activation and stops with an actionable message on older VS Code builds. The CommonJS loader also refuses a globally installed PI SDK older than 1.4.0.

How PiLot Studio relates to PI

PiLot Studio is a front end, not a fork of the agent. The extension host creates a PI AgentSession through the PI SDK and mirrors the session into a Svelte webview. Prompts, tool calls, streaming responses, session files, packages, skills, MCP servers, and credentials all use PI's own formats and storage.

That means:

A session started in the Pi CLI can be opened in PiLot Studio, and the reverse is also true.
~/.pi/agent (or your configured agent directory) remains the source of truth for auth.json, models.json, settings.json, mcp.json, sessions, packages, and trust.json.
PiLot Studio is independent of the Pi project. It is published by Printaga Publishing LLC under Apache-2.0.

What changed in 3.0.0

PiLot Studio 3.0.0 (October 24, 2026) moved to PI 1.4 and added:

MCP management. A dedicated MCP tab for global mcp.json and trusted project .pi/mcp.json, with live server state, tool lists, scope badges, exposure controls, add/remove, and OAuth sign-in.
Codemode controls. Settings for codemode mode, inline tool budget, automatic activation, and adding +codemode to the default tools.
Project trust. Project-local .pi settings, MCP servers, packages, and extensions only load after you trust the workspace. The decision is shared with the Pi CLI through trust.json.
Cache warming. A per-session control for PI's cacheWarming mode (Streaming, Idle, Off).
Media catalog visibility. Image and classifier models from PI's catalog appear in the Providers panel; chat pickers still list chat models only.
Online model refresh. An opt-in refresh that uses PI's personalized network catalog, alongside the default offline refresh.
Higher runtime floor. VS Code 1.105 and Node 22.19 are now required, matching the PI 1.4 SDK.

Where to go next

Install and first run if you have not used PiLot Studio before.
The PiLot Studio workspace for a tour of the interface.
MCP servers and codemode for the 3.0 feature set.
Troubleshooting if PI, a provider, or an MCP server is not being detected.