Your agents are
thinking.
Watch them do it.

The Agent Observability Protocol (AOP) is an open protocol for real-time observability of autonomous AI agents. Know what they're doing, why they're doing it, and exactly where things go wrong.

read the spec →
2 lines
to integrate
LLM agnostic
any framework
local first
open source
live
research-agent
step 0 · 0 tokens
00:52:59
 
 
00:52:59
 
 
00:52:59
 
 
00:52:59
 
 
00:52:59
 
 
00:52:59
 
 
cost $0.000uptime 0s
kill session
01
Add the SDK
Two lines in your agent. Works with any LLM or framework.
02
Start the collector
One command. Runs locally. No account required.
03
Watch your agent think
Open the dashboard. See every decision in real time.
integration

Two lines to full visibility.

import { AOPClient } from '@useaop/sdk'

const aop = new AOPClient({ agentId: 'my-agent' })

await aop.sessionStarted({ goal: 'Research competitors' })
await aop.thought('Starting with broad search', { confidence: 'high' })

const id = await aop.toolStart('web_search', { query: 'AI tools 2026' })
await aop.toolEnd('web_search', true, 'Found 10 results', 420, { toolCallId: id })

await aop.sessionEnded('completed', { outcome_summary: 'Research complete' })
protocol

An open standard, not a product.

AOP is a specification anyone can implement. The SDK and collector are reference implementations. Build your own collector, your own dashboard, your own integrations — the protocol is MIT licensed and belongs to the community.

read the specification →
session.startedsession.heartbeatsession.ended
cognition.thoughtcognition.goalcognition.decisioncognition.uncertainty
operation.tool_startoperation.tool_endoperation.agent_spawnoperation.memory