CARC Developer Platform

Appearance

CARC Developer PlatformIntegrate consequential AI agents with runtime control

Register an agent, submit an Agent Run, receive a control decision, retrieve the audit record — using documented APIs and an SDK, without reading engine source.

Quickstart
Overview
REST API Guide

Runtime Control, not model execution

The engine never generates agent output. It evaluates whatever your agent already produced against evidence, configurable policies, and runtime risk, then returns allow, require review, or block.

SDK or REST

A thin typed TypeScript client (@giggle-ai/runtime-sdk), or call the REST API directly with curl. Both surfaces are documented here.

Immutable audit trail

Every Agent Run produces one immutable Decision Context Record. Decision Replay reconstructs exactly what happened and why, across every retry.

Where to start ​

  1. Overview — what this platform is and isn't
  2. Core Concepts — Runtime Session, Agent Run, Decision Context Record
  3. Quickstart — a runnable, copy-paste integration in minutes
  4. Reference Integrations — three worked examples with distinct outcomes

The engine also serves an interactive, always-current reference at /docs, generated from the same OpenAPI spec this site's REST API guide is written against. Set CARC_RUNTIME_API_URL to the engine origin before following the guide's examples.

Developer Platform v1: Consequential Agent Runtime Control Platform

Not for use with real personal data. Synthetic data only.