How Monte works.

Your agent still writes the answer. Monte checks whether the answer is safe to show, before it reaches a user.

Step 01

Your agent writes an answer.

A user asks a question in your product. Your agent writes an answer. Before that answer reaches the user, your product hands it to Monte.

User
Can I get a refund after 60 days?
Agent
Yes, you can get a refund within 60 days.
Step 02

Your product sends Monte the answer.

Three things go in: the user's question, the agent's answer, and the source text or data the answer should be based on.

Question
Can I get a refund after 60 days?
Answer
Yes, you can get a refund within 60 days.
Sources
Refunds are available within 30 days of purchase.
Step 03

Monte checks the answer.

Monte looks for claims that are not supported, numbers that do not match, missing proof, risky statements, and answers that should go to a human.

Yes, you can get a refund within 60 days.
Conflict with source
The agent said 60 days. The source says refunds are available within 30 days.
Step 04

Monte returns a result.

A clear, structured response your product can act on. Confidence, risk level, flagged claims, and a recommended action.

Confidence
low0.42
Risk level
high
Flagged claim
You can get a refund within 60 days.
Source says 30 days, not 60.
Recommend
Send to human
Step 05

Your product decides what to do.

You set the rules. Monte gives your product the signal it needs.

Monte result
Product action
Safe
Send the answer
Unclear
Retry or show a warning
Risky
Block the answer
High risk
Send to a human

Where Monte fits.

Monte sits between your agent and your users. It does not replace your model. It checks the answer your model already wrote.

01Your app
02Your agent
03Monte
04Your user

Three ways to use Monte.

API
Call Monte before showing an agent answer. Best for production apps, support bots, copilots, RAG answers, and summaries.
Playground
Paste in a question, an agent answer, and source text. Monte shows what is supported and what looks risky. Best for testing, demos, and early setup.
Rules
Decide what should happen when Monte finds risk. Send safe answers, retry unclear ones, block risky ones, route high-risk to a human.

What Monte is best for.

Monte works best when an agent answer should be grounded in something real.

Grounded in
  • Help docs
  • Company policies
  • Product data
  • Account data
  • Support articles
  • Internal knowledge
  • Retrieved sources
Common formats
  • Support replies
  • Agent summaries
  • RAG answers
  • Agent copilots
  • Customer-facing chatbots
What Monte is not

Not a new agent or model. Not a chatbot. Not a replacement for your product. Monte checks the answer before your product shows it. That is the whole job.

Stop bad answers.

Monte checks every reply your agent writes and flags what isn't supported, before a user ever sees it.