Tool calls
The right function, with the right arguments.
Turn a request into the exact function call your system expects: the right tool, valid arguments, your field names and conventions. It is the part of an agent that actually does things, trained to do it the way your systems need.
Request
Maya studied at RISD, she's a set designer at the Public Theater.
Model output · real · exactly matches the label
update_contact(name="Maya", field="employer", value="Public Theater") update_contact(name="Maya", field="role", value="set designer") update_contact(name="Maya", field="school", value="RISD")
Your CRM
What it does
Built for tasks like these.
- Turn call notes into CRM updates
- Drive internal APIs from plain requests
- Fill tickets and forms from a message
- Give an agent reliable actions
How it works
From your examples to a model you own.
- 01
Collect calls that worked
Requests paired with the call your team or system made in response. The contact model learned from 801.
- 02
Train on your tools
The model learns your function names, argument formats and when to make no call at all.
- 03
Call your systems
Validate each call against your schema and run it, all inside your network.
Private by default
Your requests never leave the building.
Tool calls carry the data they act on: customer names, amounts, account details. A local model reads the request and writes the call without any of it leaving your network.
- Your customers' data leaves the building on every call
- The model changes whenever they ship
- You pay for a huge general model to do one narrow task
- The request never leaves the device
- The model is a file you own; it changes when you retrain it
- Small enough to run on hardware you already have
Trained, not prompted
Accurate on your task.
Our contact-notes model writes update_contact calls from meeting notes and gets 137 of 142 held-out notes exactly right. The best frontier model got 77; most of its misses were conventions, not comprehension.
And it costs less
No per-call bill.
Agents make many calls per task. When each step is a local model, the loop costs nothing per step.
$120,000 annual savings
See it for yourself
Run this model on your own GPU.
A guided demo, about a minute. It downloads once and runs entirely in your browser; nothing you type is sent anywhere.Open the demoRead and write
Related solutions.
Have a task like this?
We'll tell you if a micromodel can do it. Or book a demo, or email hello@micromodel.company.