One decorator.
A shareable eval report.

Drop AgentsProof into your agent in 30 seconds. Get a public URL that proves it works — scored on goal completion, tool accuracy, efficiency, quality, and safety.

$ npm install @agentsproof/sdk

agentsproof.dev/r/x8kp2m live

agent.tsyour code

import { AgentsProof } from '@agentsproof/sdk';

const ap = new AgentsProof({ apiKey: process.env.AGENTSPROOF_API_KEY });
const run = ap.startRun({ projectSlug: 'my-agent', input: { query } });

const result = await run.trace('llm_call', 'gpt-4o', () => llm(query));
const { publicUrl } = await run.complete({ answer: result });

console.log(publicUrl); // → https://agentsproof.dev/r/abc123

report · run #47the proof

my-coding-agent

2 minutes ago · 8 steps

87/100

Goal completion92

Tool accuracy78

Step efficiency95

Output quality88

Safety97

No anomalies

Drops intoOpenAIAnthropicLangChainVercel AI SDKLlamaIndex

How it works

Three lines between you and proof.

01run.install()

Install the SDK

One package, zero config. Works in any Node or edge runtime.

$ npm i @agentsproof/sdk

02run.trace()

Wrap your calls

Drop run.trace() around each LLM and tool call. That's the whole integration.

$ run.trace('llm_call', fn)

03run.share()

Get a public URL

Every run is scored on 5 axes and gets a shareable report card.

$ → agentsproof.dev/r/x8kp2m

agent.tsCopy

import { AgentsProof } from '@agentsproof/sdk';

const ap = new AgentsProof({ apiKey: process.env.AGENTSPROOF_API_KEY });
const run = ap.startRun({ projectSlug: 'my-agent', input: { query } });

const result = await run.trace('llm_call', 'gpt-4o', () => llm(query));
const { publicUrl } = await run.complete({ answer: result });

console.log(publicUrl); // → https://agentsproof.dev/r/abc123

From the timeline

Devs are posting their scores.

@nadia_builds· 2h

Shipped 4 agents this month. AgentsProof is the first time I actually know which one is good.

@theo_dev· 1d

The public report URL is clever. Posted my score on HN and got 3 people asking what I used.

@marcus_ml· 3d

5 minute setup. Caught a tool-call loop I'd never have spotted otherwise.