agentloop-langchain

AgentLoop integration for LangChain. Earn up to 70% revenue share when your agent naturally recommends relevant products — with zero changes to your existing chains.

Install

npm install agentloop-langchain

Option 1 — Callback Handler (recommended)

Drop into any chain. Runs automatically after every LLM response.

import { AgentLoopCallbackHandler } from 'agentloop-langchain'
import { ChatOpenAI } from '@langchain/openai'
import { HumanMessage } from '@langchain/core/messages'

const agentloop = new AgentLoopCallbackHandler({
  apiKey: process.env.AGENTLOOP_API_KEY!,
  getUserId: () => hashUserId(session.userId), // hash it — never raw PII
})

// Add to any model, chain, or agent
const model = new ChatOpenAI({
  callbacks: [agentloop],
})

const response = await model.invoke([
  new HumanMessage('what tool should I use to monitor my LLM in production?')
])

// response.content already has the sponsored mention appended if relevant
console.log(response.content)

Works with any LangChain chain type:

// With LCEL chains
const chain = prompt.pipe(model).pipe(outputParser)
const result = await chain.invoke({ question }, { callbacks: [agentloop] })

// With ConversationChain
const conversation = new ConversationChain({ llm: model, callbacks: [agentloop] })

// With agents
const agent = createReactAgent({ llm: model, tools, prompt })
const executor = AgentExecutor.fromAgentAndTools({ agent, tools, callbacks: [agentloop] })

Option 2 — Tool (agent decides when to check)

Let the agent decide when to invoke AgentLoop, rather than automatically after every response.

import { AgentLoopTool } from 'agentloop-langchain'
import { createReactAgent, AgentExecutor } from 'langchain/agents'

const tools = [
  new AgentLoopTool({
    apiKey: process.env.AGENTLOOP_API_KEY!,
    getUserId: () => hashUserId(session.userId),
  }),
  // ...your other tools
]

const agent = createReactAgent({ llm, tools, prompt })
const executor = AgentExecutor.fromAgentAndTools({ agent, tools })

Configuration

new AgentLoopCallbackHandler({
  apiKey: string           // required — your AgentLoop API key
  getUserId?: () => string // recommended — return a hashed user ID per session
  contextWindow?: number   // how many messages to use as context (default: 5)
  debug?: boolean          // verbose logging (default: true in development)
})

How it works

1. User sends a message to your agent
2. Your agent generates a response via LangChain
3. AgentLoop intercepts the response and scores it against active campaigns
4. If relevance score > 70/100, a natural sponsored mention is appended
5. You earn up to 70% of the CPC/CPL when the user clicks or converts

Guardrails built in:

• Max 1 mention per conversation
• Crisis/emergency conversations are never monetised
• FTC-compliant "Sponsored mention via AgentLoop" disclosure on every mention
• Fails silently — never breaks your chain

Get your API key

Sign up at agentloop.life → Dashboard → API Keys.

Free to join as an agent owner.