Phi (φ) — Pi AI Agent for VS Code

The golden ratio of AI-powered development.

Support the project by giving the repo a ⭐ and showing some ❤️!

Phi brings the Pi AI coding agent into VS Code as a native extension. Chat with an AI agent that can read, write, and edit your code — directly from the sidebar.

⚠️ Note: Phi is a community-built extension and does not yet cover the full Pi agent feature set. It's a work in progress, built through vibe coding, and we welcome all contributions to help reach full Pi parity. See Contributing below.

✨ Features

💬 AI Chat in the Sidebar

A full-featured chat interface right inside VS Code. Send messages, receive streamed responses, and watch the agent think and work in real time. Assistant code blocks are syntax-highlighted across many programming languages and follow your VS Code theme.

🛠️ Tool Execution Cards

See exactly what the agent is doing — read, write, edit, and bash tool calls render as collapsible cards with live output streaming, inline diffs for edits, copy buttons, and auto-collapse on completion.

🧠 Thinking Blocks

When the model thinks before responding, you see it. Collapsible thinking blocks show the agent's reasoning process, toggleable from settings.

📂 Editor-Aware Context

Phi knows what you're working on:

Add selection to chat — select code, press ⌘+ (Mac) / Ctrl++ (Win/Linux), or right-click → "Phi: Add to Chat"
Add file to chat — right-click a file in Explorer → "Phi: Add File to Chat"
Attach files — click the 📎 button in the input area to attach images or any file type via the native file picker
Paste images — paste images from clipboard (Cmd+V) to attach them inline
Ask about selection — right-click selected code → "Phi: Ask About Selection"

Context references appear as lightweight chips showing the filename only — Pi reads the full file content itself.

📜 Session History

Browse and switch between past conversations, grouped by project. Search sessions by name. Favorites supported. Restored chats show the full current branch and mark where context compaction happened, with expandable compaction summaries.

🌿 Conversation Tree

Navigate conversation branches, set labels on entries, and branch with optional context summaries. Full tree visualization with role icons and branch count badges.

🔧 Model & Settings

Model dropdown with search — switch between all available models
Thinking level — cycle off / low / medium / high
Auto-compaction — toggle automatic context compaction
Manual compaction — via command palette with progress indicator
Experimental task alert sounds — optional sounds for successful completion and failed runs (still under development/testing)
Manage Pi Extensions — Settings opens a dedicated extension manager where loaded Pi extensions, including the built-in legacy Google providers, can be enabled or disabled
Session cost & token usage — live display in the footer with context window visualizer

🔑 Accounts & Auth

Unified login / setup flow — subscriptions via OAuth plus API-key-capable providers discovered from Pi's model registry
Guided provider environment setup — providers can ask for required env vars step-by-step during Login / Setup
Global or Phi-local env vars — if VS Code sees an existing env var, Phi offers to use it; otherwise values can be saved locally in VS Code SecretStorage and applied only inside Phi
Phi: Add API Key remains a shortcut — for built-in and custom non-OAuth providers, but Phi: Login is now the primary entry point
Stored credentials live in ~/.phi/auth.json — separate from Pi CLI auth; environment and models.json auth still work too
Cloudflare Workers AI & AI Gateway — guided setup for required CLOUDFLARE_ACCOUNT_ID and CLOUDFLARE_GATEWAY_ID
Amazon Bedrock — guided setup for AWS profile, IAM keys, or bearer-token env vars
Built-in legacy Google providers extension — Phi keeps Google Cloud Code Assist (Gemini CLI) and Google Antigravity available via a built-in Pi extension, even though newer Pi SDK versions removed these providers after Google started restricting some external OAuth usage. Release builds can embed Google OAuth credentials at build time for Pi-like out-of-the-box login; if none are embedded, configure your own client ID/secret during Login / Setup. Use responsibly and follow Google's account terms; the extension can be disabled from Settings → Manage Pi Extensions.
No model available? — the header model control turns into a Login button that opens the Accounts panel

🖥️ Custom Providers (Ollama, vLLM, LM Studio…)

Phi inherits full custom provider support from the Pi SDK. Add any OpenAI-compatible local or remote model by editing ~/.pi/agent/models.json — no extension restart needed, changes are picked up next time you open the model picker.

Maintainer note: legacy Google OAuth defaults

To make published builds behave like Pi 0.70.6, set these as CI/local build environment secrets before packaging. For local testing, scripts/build-ext.mjs also loads them from .env. They are embedded into the generated .vsix, but never committed to source:

PHI_EMBEDDED_GOOGLE_GEMINI_CLI_OAUTH_CLIENT_ID
PHI_EMBEDDED_GOOGLE_GEMINI_CLI_OAUTH_CLIENT_SECRET
PHI_EMBEDDED_GOOGLE_ANTIGRAVITY_OAUTH_CLIENT_ID
PHI_EMBEDDED_GOOGLE_ANTIGRAVITY_OAUTH_CLIENT_SECRET

See Custom Providers for setup instructions.

⌨️ Keyboard Shortcuts

Action	Shortcut
Open Phi chat	`Cmd+Shift+L` / `Ctrl+Shift+L`
Add selection to chat	Select code → `⌘+` (Mac) / `Ctrl++` (Windows/Linux)
Abort current turn	`Escape` (when panel focused)
Focus chat input	Press `/` inside the Phi panel

🎨 Native VS Code Theming

No custom themes — Phi automatically follows your VS Code theme (dark, light, high contrast) using built-in --vscode-* CSS variables.

📦 Install

From Source (Development)

git clone https://github.com/gnassro/phi.git
cd phi
pnpm install              # or: npm install
pnpm run build            # or: npm run build

# Press F5 in VS Code to launch Extension Development Host

Package & Install Locally

# automatically runs build first
pnpm run package          # or: npm run package
code --install-extension phi-agent-0.7.2.vsix

🧰 Usage

Action	How
Open Phi chat	`Cmd+Shift+L` / `Ctrl+Shift+L`
Ask about selected code	Right-click → "Phi: Ask About Selection"
Add selection to chat	Select code → `Cmd+Shift+=`
Add file to chat	Right-click file in Explorer → "Phi: Add File to Chat"
Attach files	Click 📎 in the input area (images + any file type)
Paste images	`Cmd+V` / `Ctrl+V` with image in clipboard
New session	Command Palette → "Phi: New Session"
Switch session	Click the 🕐 history button → select a session
Switch model	Click the model dropdown in the header
Compact context	Click the commands button (in chat input) → "Compact"
View session stats	Click the commands button (in chat input) → "Session Stats"
Login / provider setup	Command Palette → "Phi: Login" or the Accounts panel button
Add API key (direct shortcut)	Command Palette → "Phi: Add API Key"
Add custom provider	Edit `~/.pi/agent/models.json` (see Custom Providers)

⚠️ Disclaimer

This extension is in early development and comes with no warranty. Please be aware:

Not all features have been fully tested — some may behave unexpectedly
Only tested on macOS — Windows and Linux are untested and may have issues
Use at your own risk — always review AI-generated code changes before accepting them

If you encounter any bugs or issues, please open an issue on GitHub. Your feedback helps improve the extension for everyone.

🖥️ Custom Providers

Phi supports any OpenAI-compatible model server (Ollama, vLLM, LM Studio, OpenRouter, proxies, etc.) through the Pi SDK's models.json config file. No code changes, no extension restart — just edit the file and open the model picker.

Setup

1. Create or edit ~/.pi/agent/models.json

This file is shared with the Pi CLI, so any provider you add works in both.

2. Add your provider

{
  "providers": {
    "ollama": {
      "baseUrl": "http://localhost:11434/v1",
      "apiKey": "OLLAMA_API_KEY",
      "api": "openai-completions",
      "compat": {
        "supportsDeveloperRole": false,
        "supportsReasoningEffort": false,
        "maxTokensField": "max_tokens"
      },
      "models": [
        {
          "id": "llama3.1:8b",
          "name": "Llama 3.1 8B (Local)",
          "reasoning": false,
          "input": ["text"],
          "cost": { "input": 0, "output": 0, "cacheRead": 0, "cacheWrite": 0 },
          "contextWindow": 128000,
          "maxTokens": 32000
        }
      ]
    }
  }
}

3. Open the model picker in Phi and select your model

The new model will appear immediately — no restart needed.

The `compat` block (important for local models)

Most local servers (Ollama, vLLM, LM Studio, SGLang) don't fully implement the OpenAI spec. Without the compat block, the system prompt gets sent using the developer role which local servers silently ignore — causing the model to act as plain chat instead of a coding agent.

Flag	What it fixes
`supportsDeveloperRole: false`	Sends system prompt as `system` role (understood by all servers)
`supportsReasoningEffort: false`	Disables `reasoning_effort` param (unsupported by local servers)
`maxTokensField: "max_tokens"`	Uses `max_tokens` instead of `max_completion_tokens`

Always include the compat block for Ollama, vLLM, LM Studio, and similar servers.

Supported API types

`api` value	Use for
`openai-completions`	Ollama, vLLM, LM Studio, OpenRouter, most compatible servers
`anthropic-messages`	Anthropic Claude API or compatible proxies
`openai-responses`	OpenAI Responses API
`google-generative-ai`	Google Gemini API

Authentication for custom providers

Phi's Add API Key dialog only manages built-in providers. For custom providers, configure authentication directly in ~/.pi/agent/models.json via the provider's apiKey field.

The apiKey value can be either:

an environment variable name such as OPENROUTER_API_KEY
a literal value for local or private setups

For local servers like Ollama that don't require a real key, set "apiKey": "ollama" in models.json (any non-empty value works).

Multiple providers

You can define as many providers as you like:

{
  "providers": {
    "ollama": { ... },
    "lm-studio": {
      "baseUrl": "http://localhost:1234/v1",
      "apiKey": "lm-studio",
      "api": "openai-completions",
      "compat": {
        "supportsDeveloperRole": false,
        "supportsReasoningEffort": false,
        "maxTokensField": "max_tokens"
      },
      "models": [
        { "id": "qwen2.5-coder-32b", "name": "Qwen 2.5 Coder 32B", "reasoning": false, "input": ["text"], "cost": { "input": 0, "output": 0, "cacheRead": 0, "cacheWrite": 0 }, "contextWindow": 128000, "maxTokens": 32000 }
      ]
    },
    "openrouter": {
      "baseUrl": "https://openrouter.ai/api/v1",
      "apiKey": "OPENROUTER_API_KEY",
      "api": "openai-completions",
      "models": [
        { "id": "meta-llama/llama-3.1-8b-instruct", "name": "Llama 3.1 8B (OpenRouter)", "reasoning": false, "input": ["text"], "cost": { "input": 0.1, "output": 0.1, "cacheRead": 0, "cacheWrite": 0 }, "contextWindow": 131072, "maxTokens": 8192 }
      ]
    }
  }
}

For the full models.json reference, see the Pi SDK documentation.

🏗️ Architecture

Phi is a VS Code extension built with TypeScript + vanilla JS:

Extension Host (Node.js) — runs the Pi SDK directly, manages sessions, handles auth
Webview (Chromium sandbox) — chat UI, tool cards, settings panels
IPC — all communication via VS Code's built-in message passing (postMessage)

The Pi SDK runs in the same Node.js process as the extension host — no external servers, no WebSocket, no HTTP. Sessions are stored at ~/.pi/agent/sessions/ (shared with the Pi CLI).

🔗 Relation to Pi

Project	What it is
Pi	The CLI AI coding agent (`pi` command)
Phi	A VS Code extension that brings Pi into the editor

Phi uses the Pi SDK (@mariozechner/pi-coding-agent@0.73.0) to run the agent directly inside VS Code's extension host.

Pi SDK compatibility: Phi is built and tested against Pi SDK 0.73.0. Newer versions may work but are not guaranteed until tested.

🚧 Current Status

Phi is functional and covers the core Pi agent experience, but it is not yet a complete implementation of all Pi features. Here's what's included:

✅ What Works

Full chat with streaming responses
Tool execution (read, write, edit, bash) with live output
Session history, switching, and continuity
Conversation tree with branching and navigation
Model switching, thinking levels, context compaction
Unified subscription login/setup + API key management
Editor context integration (selection, file, diagnostics)
Cost and token tracking with context window visualizer

We welcome contributions to add more features! See Contributing.

🤝 Contributing

Phi is a vibe-coded project — built with AI agents (using Pi), iteratively refined, and open to everyone.

Whether you want to fix a bug, add a missing Pi feature, improve the UI, or write tests — contributions are welcome!

How to Contribute

Fork the repository
Clone and install: pnpm install
Build: pnpm run build
Test: Press F5 in VS Code to launch the Extension Development Host
Submit a PR with a clear description of what you changed and why

Areas Where Help Is Needed

Bringing more Pi agent features into Phi
Testing on different platforms (Windows, Linux)
UI/UX improvements
Documentation
Accessibility

See AGENTS.md for the full technical reference — it's written for both humans and AI agents working on this codebase.

📝 License

This project is licensed under the MIT License.

🙏 Credits

Built on top of the Pi agent by Mario Zechner.

Name		Name	Last commit message	Last commit date
Latest commit History 101 Commits
.github		.github
.pi/skills/phi-developer		.pi/skills/phi-developer
assets		assets
docs		docs
public		public
scripts		scripts
src		src
.gitignore		.gitignore
.vscodeignore		.vscodeignore
AGENTS.md		AGENTS.md
CHANGELOG.md		CHANGELOG.md
LICENSE		LICENSE
README.md		README.md
package.json		package.json
pnpm-lock.yaml		pnpm-lock.yaml
tsconfig.json		tsconfig.json

Uh oh!

Folders and files

Latest commit

History

Repository files navigation

Phi (φ) — Pi AI Agent for VS Code

✨ Features

💬 AI Chat in the Sidebar

🛠️ Tool Execution Cards

🧠 Thinking Blocks

📂 Editor-Aware Context

📜 Session History

🌿 Conversation Tree

🔧 Model & Settings

🔑 Accounts & Auth

🖥️ Custom Providers (Ollama, vLLM, LM Studio…)

Maintainer note: legacy Google OAuth defaults

⌨️ Keyboard Shortcuts

🎨 Native VS Code Theming

📦 Install

From Source (Development)

Package & Install Locally

🧰 Usage

⚠️ Disclaimer

🖥️ Custom Providers

Setup

The compat block (important for local models)

Supported API types

Authentication for custom providers

Multiple providers

🏗️ Architecture

🔗 Relation to Pi

🚧 Current Status

✅ What Works

🤝 Contributing

How to Contribute

Areas Where Help Is Needed

📝 License

🙏 Credits

About

Topics

Resources

License

Uh oh!

Stars

Watchers

Forks

Releases 11

Sponsor this project

Uh oh!

Packages 0

Uh oh!

Contributors

Uh oh!

Languages

The `compat` block (important for local models)

Packages