arc-guardrails

Appearance

arc-guardrailsGuardrails for production LLM systems

A protocol-driven Python SDK and service surface for detecting, transforming, routing, and auditing risky AI traffic.

Read the guide
Explore canvases
View reference

Protocol-first extension

Extend inspectors, strategies, lifecycle sinks, and policy surfaces without subclassing framework internals.

End-to-end request control

Run validation, detection, sanitization, routing, refusal, verification, and reporting as one coherent pipeline.

Transport-neutral deployment

Use the library in-process, through a service boundary, or alongside the operator dashboard and local stack.

Three packages · one decision contract · one local developer surface

arc-guardrails is an open source guardrails project for teams that need to keep sensitive or adversarial content under control before it reaches downstream language models. The repository packages the contract layer, the batteries-included runtime, and the service layer separately so operators can adopt only the surface they need.

Start Here ​

Guide

Understand the package split, the 12-stage execution model, and the local setup path that the Makefile already standardizes.

Open section ->

Reference

Trace the pipeline, the API surface, observability hooks, and the detection coverage matrix from the existing knowledge base.

Open section ->

Canvases

Use interactive request and pipeline flow views derived from the repository’s existing canvas documents.

Open section ->

Run Locally

Use the repository Makefile to install packages, boot the API, start Docker services, and exercise the stack on any machine.

Open section ->

Reading Paths ​

  • New to the project: start with Guide, then move to the architecture and pipeline reference pages.
  • Evaluating integration fit: jump to Reference for packages, API surface, and technology inventory.
  • Explaining flows to operators or contributors: use Canvases for request lifecycles and branch behavior.

Open source but Use responsibly 🛡️