Quickstart
Instrument your first endpoint in under five minutes
How It Works
Understand the protocol flow from response to outcome review
SDK Guides
Node, Python, Go, Rust, and MCP integration guides
API Reference
Explore the public endpoints for outcomes and telemetry
Why Agent Feedback?
AI agents calling your product make decisions independently. Traditional analytics tell you what was called — Agent Feedback tells you whether it actually worked. When an agent completes a task, it reports a concise outcome (success, partial, or failure) along with a single explanatory sentence. That signal flows directly into your dashboard, scoped to your workspace and product.
Zero Latency
Capability tokens are signed locally — your product response never waits for a network call
Privacy First
Only outcome and a short note are collected — no prompts, transcripts, or personal data
Protocol First
One canonical signing algorithm works across every language and framework
Supported Integrations
Node — Express, Fastify, MCP
One global middleware for Express or Fastify;
instrumentMcp for MCP serversPython — ASGI & WSGI
Dependency-free middleware for FastAPI, Starlette, Flask, and Django
Go — net/http
Standard-library middleware using only built-in packages
Rust — Axum & Tower
Tower layer for Axum with a bounded Tokio background queue
Get Started in Three Steps
1
Create a product in the dashboard
Sign in, create a workspace, and add your first product to get an API key in the format
af_live_....2
Add the middleware
Install the SDK for your framework and add one line of middleware — no handler changes required.
3
Watch outcomes arrive
Deploy and let agents call your product. Outcome reviews appear in your dashboard as agents report results.