TOOL USE

Teach the model to call your APIs

Describe your private API as a skill, fine-tune the model on when and how to call it, and Harness bounds every call with scopes and an audit trail.

Skilllookup_order
Input
order_id: string
Calls
GET /orders/{id}
Scope
read
{
  "type": "tool_use",
  "name": "lookup_order",
  "input": { "order_id": "SO-4471" }
}

# the model asked; Harness checked the key's scope

Skills carry the contract

A skill definition holds the endpoint, the authentication, the parameters and the schema, so the model knows what it may call and what the call returns.

Fine-tuning teaches judgement

The model learns from examples when to reach for a skill and how to populate its parameters, so tool calls follow the patterns your data carries.

Harness bounds every call

The same four gates that govern every request check the scope on the key before a tool call executes, and the call lands in the audit trail.

How it works

From skill to tool call

A skill definition carries the HTTP method, the endpoint, the authentication pattern, the parameters with their types and the response schema. The model reads the definition at inference time, so adding a skill is a data change with no retraining.

Read the docs
Results

Results from LLMs powered by Nucleus

$73.4K

monthly revenue re-attributed in marketing

47 vehicles

rerouted in a single fleet optimisation

$177K

of downtime prevented in manufacturing

94%

prediction accuracy on equipment failures

USE CASES

Models that call the tools your teams use

Fleet operations

47 trucks rerouted while they were still moving

$2.4Kdaily fuel saved
23delays resolved without a call
IT operations

Every incident past its four-hour SLA, escalated by name

8incidents past their four-hour SLA
6.2 hrsaverage time to resolution
Manufacturing

Line 3's next failure, booked into the maintenance window

$177Kdowntime prevented this month
18.3 daysadvance notice on a failure

What tool use includes

Skills

  • Define skills as structured JSON
  • HTTP method, endpoint, auth, parameters and schema
  • Add or update a skill with no retraining
  • Response schema validates what the model returns

Training

  • Fine-tune on tool-call examples in JSONL
  • Positive and negative examples teach judgement
  • Same training meters, same base model pricing
  • 25 open base models across six modalities

Governance

  • Harness checks the infer scope on every call
  • Project-bound keys limit blast radius
  • Every tool call in the append-only audit trail
  • Four gates: organisation, authority, project, allowed
Pricing

Two meters, usage-based

Training and inference tokens

From $0.22 / 1M tokensper million tokens processed

Fine-tuning jobs from $0.22 / 1M tokens. Inference from $0.15 / 1M tokens. See pricing for the full model catalogue.

Deploy

Pick your path

Harness

Permissions, scoped keys and audit

Explore Harness

On-Premise Deployment

Our GPU servers, your site

Explore On-Premise Deployment

MCP

Your model in your tools

Explore MCP

RAG

Answers from your documents

Explore RAG

Webhooks

Signed events to your systems

Explore Webhooks

Teach the model to call your APIs.

Describe a skill, fine-tune on examples, and let Harness govern every call.