Your data, your rules, any model
Monk runs your AR through a controlled harness. It routes each task to the right model, checks every step before it acts and keeps your data inside your environment.
Your systems and rules, wrapped around the models
Your ERP, billing, bank, email, and CRM feed Monk. Agents run collections, disputes, cash application, portal submission, and forecasting through evaluations, guardrails, a model router, and tools. Your data never touches a raw model.
The best model for each task
Monk routes each task to the model that handles it best, whether that is document extraction, collections, or cash application. You are never locked to one provider. Every new frontier model is evaluated against your accounts first.
Your data stays in your environment
Monk never trains a model and never leaves your environment without permission. A deterministic check runs before anything reaches a customer. The infrastructure is SLA-backed, monitored, and SOC 2 compliant.
Frequently asked questions
What models does Monk use?
Monk routes to the best model per task across providers like Anthropic, OpenAI, and Google, and adds new frontier models as they arrive.
Does my data train the models?
No. Your data stays inside your environment and is not used to train Monk's systems or any provider's models. The harness decides what the model sees: the fields a task needs, not your ledger.
Is Monk SOC 2 compliant?
Yes. Monk is SOC 2 Type II compliant. Request the report at trust.monk.com.
What is the agent harness?
The controlled execution layer behind every agent action. It routes each task to the best model, checks every step before it acts, and logs the full trail. Agents act on your AR without your data going straight to a model.
How do you prevent mistakes that reach a customer?
A deterministic check sits between the agent and the customer: every outbound message is validated against your rules before it sends, and anything uncertain is drafted for review instead of shipped.
Put AI on your AR, safely
Your data, your rules, any model