Your agents have access to real money, real accounts, and real credentials. They can make purchases, initiate transactions, and spend on your behalf — and most of them have no ceiling, no oversight, and no way to ask permission before they act.
Monte is your agents' responsible adult. The authorisation layer between your agents and your money. Three lines of code.
import monte from 'monte-sdk'
const agent = monte.wrap(openai, {
apiKey: 'mk_...',
limit: 50 // hard limit in USD, per session
} )
await agent.chat.completions.create({ ... })THE PROBLEM
Agents are no longer just answering questions. They are booking travel, placing orders, calling APIs, and spending money — autonomously, at speed, and in most deployments without any layer between them and the consequences. When something goes wrong, the action has already been taken. The money is already spent. The transaction has already gone through. Monte enforces the boundary before any of that happens — not after.
Without Monte
With Monte
HOW IT WORKS
Monte sits between your agents and the world. It does not touch your money, your accounts, or your credentials. It enforces the rules you set before anything moves.
I
One import. Call monte.wrap() around your existing client. Your code stays exactly the same. Monte intercepts every action before it leaves your system.
II
Tell Monte what a single agent session is allowed to do and how much it is allowed to spend. That boundary is enforced on every action, before it is taken.
III
Every action logged. Every blocked event flagged. A complete record in plain English, readable by anyone on your team — not just the developer who built the agent.
USE CASES
One misinterpreted instruction or retry chain that goes too far — and you have a customer relations problem, not just an API bill. Monte enforces the boundary per session.
Agents that place orders, book travel, or manage procurement have access to real spend. Monte ensures they act within the limits you set — before the purchase confirmation arrives.
Research agents are the most likely to spiral. They retry failed queries and generate follow-up tasks that compound. Monte stops the spiral before it becomes a problem.
Agents that touch payments, invoices, or financial systems need a layer of oversight your business logic alone cannot provide. Monte is that layer — without ever touching the funds itself.
Coding agents that hit an error and retry indefinitely are among the most common sources of runaway API spend. Monte blocks the call before the limit is breached.
Long-running pipelines that call LLMs at every step are invisible until they are not. Monte gives you per-agent visibility into what each pipeline is spending, in real time.
WHAT YOU GET
Set the maximum a single agent session can spend or action. Monte enforces it before the action is taken, not after. The limit is the limit.
When an agent approaches its limit, you get an alert. When it reaches the limit, the action is blocked before it executes — not after the response comes back.
Monte blocks any agent action that would exceed the boundaries you set — API costs, transaction limits, or spend thresholds. Without ever touching your accounts or credentials.
Every action logged with a timestamp, cost, and status. Readable by a non-developer. Useful in a postmortem. Not raw JSON — plain English.
Works with OpenAI, Anthropic, Google, and more. One SDK. Every agent. No matter which model it runs on or which framework it was built with.
Not account-level. Per agent, per session. Know exactly which agent spent what, when it was stopped, and what it was attempting when Monte intervened.
HOW MONTE WORKS
Monte does not hold funds. Monte does not process payments. Monte does not have access to your bank accounts, your Stripe keys, or your corporate card.
Monte enforces rules — it does not execute transactions. When an agent attempts an action that exceeds the boundary you set, Monte blocks the action before it happens. The money never moves because the instruction never goes out.
You stay in control. Monte just makes sure your agents do too.
Monte is in early access. Free access in exchange for 20 minutes of your time — a conversation about how you are building with agents and what Monte could do better.
No procurement process. No security review. No architecture committee. Three lines of code and your first protected agent.
Get started freeWorks with OpenAI, Anthropic, and Google models.