tokenhush documentation

tokenhush is a local, loopback-only gateway that sits between your AI tools and their providers. It redacts request bodies on the way out and restores them on the way back, so credentials and other sensitive values never leave your machine in the clear. These pages are the open-source documentation that ships with the core repository.

Getting started

Install a tool, prove redaction locally, then run tokenhush as a service.

• Tool setup and request routing — point any OpenAI-compatible or Anthropic client at the loopback gateway, with per-tool settings and the config reference.
• Verify redaction with a local echo upstream — prove the full round trip without contacting a real provider.
• Deployment, install, and platforms — build or install the binary, and learn where it keeps config and data.
• Configuration reference — the strict tokenhush.yaml schema: every key, its default, and upstream routing.

Concepts

How the core is put together and how to extend it.

• Architecture: layers and rule abstraction — the layering graph, the single rule abstraction, and the frozen byte surfaces.
• Plugins and the Rule extension point — the one compile-time extension point and the constraints on signed rule packs.

Security

The invariants tokenhush keeps and everything it can send off-box.

• Security model and invariants — the eight invariants, the nine residual risks, and the tests that pin them.
• Network egress disclosure — every vendor-bound request tokenhush can make.

Start from install

New here? Begin with Tool setup and request routing to install tokenhush and point your first client at the local gateway.