Does Phantom slow down my AI requests?

About 0.5 ms of proxy overhead per request — not measurable in practice. The proxy is a Rust HTTP server bound to 127.0.0.1.

What does AI see when Phantom is installed?

The .env file contains phm_xxxxxxxx tokens instead of real values. AI tools read those tokens. The local proxy swaps them for real keys just before the outbound TLS connection.

What happens if a phm_ token leaks from AI logs?

Nothing. phm_ tokens are session-scoped placeholders that have no value outside your local proxy. The real key never left your machine.

How are real keys stored?

OS keychain on macOS and Linux (Keychain / Secret Service). Encrypted file fallback for CI and Docker, using ChaCha20-Poly1305 with Argon2id key derivation.

Which editors does Phantom work with?

Claude Code, Cursor, Windsurf, Codex via MCP. Any tool that reads .env files works automatically because Phantom rewrites the file.

Is Phantom open source?

Yes. MIT licensed. Source at github.com/ashlrai/phantom-secrets. Rust workspace — phantom-core, phantom-vault, phantom-proxy, phantom-cli, phantom-mcp.

Phantom — Delegate everything to AI

Name: Phantom
Availability: InStock
Author: AshlrAI

Checking session…