Guide
The Pi Coding Agent for VS Code
Pi is an open-source AI coding agent that runs in your terminal: it reads and edits your code, runs commands, and works with the models you choose. PiLot Studio brings that same agent into VS Code as a first-class graphical interface — so you can run Pi in VS Code with a chat panel, session management, and point-and-click control instead of memorized commands.
What is the Pi coding agent?
Pi (installed as the pi CLI from pi.dev) is a minimal,
multi-model coding agent. Unlike cloud-locked assistants, Pi runs locally: it connects
to Anthropic, OpenAI, Google, Ollama, and OpenAI-compatible providers with
your own API keys, keeps sessions on disk, and is extensible through community packages, skills,
and prompt templates — with built-in MCP servers, codemode, and tool search since Pi 1.4.
The terminal is a power tool — but it is also a context switch. Reading diffs, retyping commands, and managing sessions in a scrollback buffer takes you out of your editor.
Why run Pi inside VS Code?
- Stay in flow. The agent lives in a side panel next to your code — no alt-tabbing to a terminal.
- See the conversation. Rendered markdown, syntax-highlighted code blocks, tables, and Mermaid diagrams instead of raw ANSI output.
- Use your editor's context. Selection, diagnostics, git status, and @-mentioned files attach automatically.
- Click, don't type. Install packages, switch models, adjust thinking levels, and restrict tools from the GUI.
PiLot Studio: a GUI, not a wrapper
PiLot Studio does not re-implement the Pi coding agent — it drives the real pi CLI through its SDK. That means everything the agent can do, the GUI can do:
- Sessions as a tree — branch, fork, rename, resume, search (Ctrl+F), and export.
- Model cycling — hop between Claude, GPT, Gemini, or local models mid-session; thinking levels from off to max.
- Tool presets — default, review (read-only), none, or fully custom per session.
- Packages & skills — browse, install, disable, and remove without touching a terminal.
- MCP servers — a dedicated tab for Pi's global and project
mcp.json: connection state, tools, exposure, and OAuth sign-in. - Light Mode — a reduced agent profile for local llama.cpp models.
- System Prompt tab — inspect the exact, live prompt the model receives for the active session (SYSTEM.md, settings overrides, extension modifications), with copy-to-clipboard and word wrap.
- Voice dictation — offline whisper.cpp speech-to-text.
- Git — one-click AI commit messages from staged changes.
How it compares to running Pi in the terminal
| Pi CLI (terminal) | PiLot Studio (VS Code) | |
|---|---|---|
| Interface | Command line | Chat panel, trees, buttons |
| Editor context | Manual | Auto: selection, diagnostics, git |
| Diffs & code blocks | Text output | Apply to Editor, Preview Diff |
| Package management | CLI commands | Visual manager |
| Model/provider control | Config files | Dropdowns + OAuth login |
| Underlying agent | The same Pi coding agent, your same keys | |
Get started
Install the pi CLI (1.4.0+), add PiLot Studio from Open VSX or the VS Code Marketplace, and press Ctrl+Shift+Alt+P. PiLot Studio 3.0 requires VS Code 1.105 or newer. The
full walkthrough is in the install guide, and the GUI tour continues in Pi coding agent GUI.