AI inference for your coding agents

Connect your agents to Phantom AI. Use routing to choose cheaper models when appropriate, or select models directly. Manage funds and pay for inference from your terminal.

npm i -g @connortessaro/pai
pai login
pai workspace fund --amount 10 --card
pai codex
Copy

Connect an account or guest workspace by device approval. Run Claude Code, Codex or pi through Phantom AI with temporary configuration, then see what it spent. pai workspace status shows shared funds and recent usage.

Connect your existing agents

pai login
pai setup
Copy

Pai previews changes, asks before applying them, then connects supported agents using your saved key. Restart your agent or open a new terminal afterward. pai setup --dry-run previews without writing; pai setup --restore restores backups. Setup for each agent covers automatic setup, Cursor's manual steps and where the key goes.

For a temporary connection, use pai codex, pai claude or pai pi.

Agent skill

Install Phantom AI instructions for your coding agent through Vercel Skills:

npx skills add connortessaro/pai --skill phantom-ai
Copy

Read the skill and installation guide. This installs instructions; install the pai CLI separately with the quickstart above.

Start here

• Run agents and choose how you pay: workspace keys, MPP sessions, x402 funding and spending controls.
• Routing: choose models for auto and test a policy for free.
• Saved secrets: private key storage and the macOS Keychain.

Advanced use

• Which key a command runs as
• Direct-key purchases
• Receipts and verify
• The API address

Reference

Interface reference covers every command, MCP tool, HTTP call and environment variable. The code reference below covers public exports. Use pai --help for the main workflows and pai --help --all for all commands.

Source, issues and releases.