pi-packages

Work in progress. These packages are under active development. Configs, APIs, and formats may change at any time without backward compatibility or deprecation notices.

Extension packages for pi.

Packages

Package	Description
pi-qq	Quick questions with /qq or /btw — side-channel LLM interaction
pi-compaxxt	Enhanced session compaction with file importance heuristics
pi-vertex	Google Vertex AI provider — Gemini, Claude, and all MaaS models
pi-handoff	Context management for agentic coding workflows
pi-image-gen	Provider-agnostic image generation
pi-dynamic-models	Dynamic model discovery from any configured API server at startup
pi-huddle	Huddle mode — read-only exploration + ask_user structured elicitation
pi-plugins	Claude plugin marketplace support for Pi, including GitHub HTTPS and SSH marketplace repos
pi-multi-pass	Multiple OAuth subscriptions with rate-limit rotation, pools, and failover
pi-delegate	Delegation and phased todos: durable child runs, plus an ACP backend (delegate backend:"acp") for Codex, Claude, Amp, OpenCode and Pi agents. Replaces the retired pi-strings package

Installation

pi install @ssweens/pi-qq
pi install @ssweens/pi-compaxxt
pi install @ssweens/pi-vertex
pi install @ssweens/pi-handoff
pi install @ssweens/pi-image-gen
pi install @ssweens/pi-dynamic-models
pi install @ssweens/pi-huddle
pi install @ssweens/pi-plugins
pi install @ssweens/pi-multi-pass
pi install @ssweens/pi-delegate

Or from a local checkout:

pi install /path/to/pi-packages/pi-vertex

Screenshots

Each package with a TUI component includes a screenshot.png captured from a live pi session. To regenerate:

# All packages
./scripts/capture-all.sh

# Single package
./scripts/capture-all.sh pi-vertex

Requirements

• iTerm2 — screenshots are captured as native window captures via screencapture -l
• Google Cloud credentials — needed for pi-vertex models to appear in the selector

Configuration

Set these environment variables (or edit the defaults in scripts/capture-screenshot.sh):

export GOOGLE_APPLICATION_CREDENTIALS=/path/to/service-account.json
export GOOGLE_CLOUD_PROJECT=your-project-id
export GOOGLE_CLOUD_LOCATION=global

Adding a screenshot for a new package

1. Create a keystroke file at scripts/keystrokes/<package-name-without-pi-prefix>.sh
2. The script receives $WID (iTerm window ID) and should send keystrokes to reach the desired TUI state
3. Run ./scripts/capture-all.sh pi-<name>

Example keystroke file (scripts/keystrokes/vertex.sh):

# Open model selector (Ctrl+L), filter to vertex models
osascript -e "tell application \"iTerm2\" to tell current session of window id ${WID} to write text (ASCII character 12)"
sleep 3
osascript -e "tell application \"iTerm2\" to tell current session of window id ${WID} to write text \"ver\" without newline"
sleep 2

How it works

1. Opens a new iTerm2 window
2. Launches pi --no-session --model k2p5
3. Sends "Hello!" and waits for a response (conversation context)
4. Runs the keystroke file to navigate to the target TUI state
5. Captures the window with screencapture -l <window-id>
6. Closes the window

License

MIT