agentbill

Usage-based billing for AI agents — preflight budget guardrails in 3 lines.

npm install agentbill

Quick start

import { AgentBill } from 'agentbill'

const bill = new AgentBill({ apiKey: process.env.AGENTBILL_API_KEY })

// Before running the agent — check if the customer has budget
const check = await bill.preflight({ customerId: 'user_123', agentId: 'research', estimatedUnits: 10 })
if (!check.approved) {
  throw new Error('Budget exceeded')
}

// After the agent finishes — record what was used
await bill.record({ customerId: 'user_123', agentId: 'research', units: 8 })

Why AgentBill?

Monthly caps let agents burn through a budget in hours. AgentBill adds a preflight check — the agent asks permission before it runs, not after it's already spent the money.

• Preflight blocks the run before it starts if the customer is over budget
• Per-request ceiling: block any single run that would cost too much
• Idempotent recording: safe to call from retried or parallel workflows
• Free tier: 1,000 units/customer, no credit card required

API

new AgentBill({ apiKey })

Get your API key at agentbill.fly.dev/register.

bill.preflight(options)

Check if a customer has budget before starting work.

| Option | Type | Default | Description | |--------|------|---------|-------------| | customerId | string | "default" | Your internal customer ID | | agentId | string | required | Agent or task type identifier | | estimatedUnits | number | 1 | Expected units for this run | | ceiling | number | — | Block run if estimatedUnits exceeds this |

Returns { approved: boolean, remainingUnits: number \| null }.

bill.record(options)

Record a billable event after work completes.

| Option | Type | Default | Description | |--------|------|---------|-------------| | customerId | string | "default" | Your internal customer ID | | agentId | string | required | Agent or task type identifier | | units | number | 1 | Units consumed | | metadata | object | — | Key-value pairs stored with the event |

Returns { recorded: boolean, eventId: string, remainingUnits: number \| null }.

LangChain integration

import { AgentBill } from 'agentbill'
import { ChatAnthropic } from '@langchain/anthropic'
import { createReactAgent } from '@langchain/langgraph/prebuilt'

const bill = new AgentBill({ apiKey: process.env.AGENTBILL_API_KEY })

async function runAgent(customerId: string, input: string) {
  const check = await bill.preflight({ customerId, agentId: 'assistant', estimatedUnits: 5 })
  if (!check.approved) throw new Error(`Blocked: ${check.remainingUnits} units remaining`)

  const agent = createReactAgent({ llm: new ChatAnthropic({ model: 'claude-3-5-haiku-latest' }), tools: [] })
  const result = await agent.invoke({ messages: [{ role: 'user', content: input }] })

  await bill.record({ customerId, agentId: 'assistant', units: 5 })
  return result
}

Links

• Dashboard
• Full docs
• GitHub
• Python SDK (PyPI)