⚡ Local Proxy · Telemetry Evidence Layer · Linux / WSL2

See what your
AI CLI sends.

Your AI coding tools send more than your prompts. Frank Infinity sits between supported AI CLIs and their upstream APIs. It separates real model traffic from telemetry, blocks or stubs known tracking, strips fingerprinting headers, and captures an evidence log of what actually left your machine.

Get Frank → $7 How it works

What Frank Does

⚡

Traffic classification

Separates AI request traffic from telemetry and other side-channel traffic. You see what's model-related and what isn't.

🧟

Telemetry blocking

Blocks or stubs known telemetry endpoints by domain, route shape, body shape, and tracking query parameters. Default preset is shield.

🧪

Evidence capture

Logs what was sent, what was blocked, and what still passed through. Surfaces it in a Textual TUI and CLI reports.

⚗️

Header stripping

Strips common tracking and fingerprinting headers from allowed AI requests. Reduces passive identifier leakage.

See It Running

Real session. Ghost preset active. Two CLIs intercepted, 42 telemetry events blocked.

Frank TUI overview dashboard
Overview — operator dashboard
Proxy targets list
Targets — per-CLI port list
Live traffic waterfall
Live — traffic waterfall, 30 events
Blocked telemetry summary
Telemetry — 42 blocked, 102 patterns
Data collected by AI tools
Collected — what companies see

Supported CLI Profiles

Each CLI gets its own dedicated provider adapter. Keep your existing workflow — Frank intercepts the traffic before it reaches the provider.

Claude OpenCode Gemini Codex Kimi KiloCode aider Hermes

Presets

🛡

shield recommended

The default and most reliable preset. Blocks or stubs matched telemetry, strips tracking headers, aims to preserve normal CLI use.

👻

ghost experimental

More aggressive blocking mode. Still experimental on some provider paths. Use only if you're comfortable debugging broken requests.

📡

static experimental

Most aggressive stubbing mode. Not universally safe. For advanced users running controlled tests.

What You Get

# Quick install
python3 -m venv .venv
source .venv/bin/activate
pip install -U pip
pip install -e .

# Launch the dashboard
frank-tui

# Start a proxy
frank start --cli opencode --preset shield

Platform Support

✓ Linux — recommended, primary target
~ Windows via WSL2 — works, Linux-first docs apply
✗ Native Windows — not ready yet
? macOS — not documented or tested; may work but not supported

Honest Limitations

○Coverage is signature-based. Frank blocks what it knows about — it does not guarantee exhaustive telemetry elimination.
○Frank is not an anonymity tool. It reduces telemetry and strips identifiers. It does not make you anonymous.
○Provider support is uneven. Some CLIs and presets work better than others. shield is the reliable baseline.
○This is source code, not an installer. Setup requires Python 3, a venv, and basic CLI comfort.
○Telemetry mutation (ghost, static) is less mature than blocking and stubbing. Start with shield.
○TLS inspection requires certificate trust installation. Some tools with cert pinning can't be fully inspected.

Get Frank

$7

Early operator pricing. One-time purchase, no subscription. Priced low while Frank is still maturing. Buying now gets you the full current release and supports continued hardening, compatibility work, and telemetry research. The price will go up as the tooling does.

Why $7? Because I want it in the hands of people who will actually run it against real CLIs — not priced for a market that hasn't formed yet.

Buy on Gumroad → $7

Non-commercial source-available license. You can use, study, and modify it. You cannot resell it.