Skip to main content

Documentation Index

Fetch the complete documentation index at: https://docs.complyhat.ai/llms.txt

Use this file to discover all available pages before exploring further.

Before you can run bias tests, generate compliance reports, or set up drift monitors, you need a registered model. The models tool is the root entity in ComplyHat — every other tool takes a model_id that traces back to a record you create here. Registration takes one call and returns a stable mdl_... identifier you reuse across your entire compliance workflow.
1

Register the model

Call the models tool with mode: "register". Provide at minimum a name, model_type, use_case, and risk_tier. The response includes the auto-generated model ID and an initial compliance_status of not_assessed.
{
  "tool": "models",
  "arguments": {
    "mode": "register",
    "name": "fraud-detector-v2",
    "version": "2.0.0",
    "model_type": "classification",
    "use_case": "fraud detection",
    "risk_tier": "high",
    "description": "Gradient-boosted classifier for real-time payment fraud scoring.",
    "owner_name": "Risk Engineering"
  }
}
The response contains an id field in the format mdl_.... Save this value — you will pass it to every subsequent tool call for this model.
description, version, and owner_name are optional but recommended. Audit reviewers and generated reports surface these fields, and omitting them produces [AMBIGUOUS]-tagged sections in reports.
2

Verify the registration

Confirm the model was stored correctly by retrieving it with mode: "get".
{
  "tool": "models",
  "arguments": {
    "mode": "get",
    "model_id": "mdl_01j9z..."
  }
}
The response echoes all fields plus the created_at and updated_at timestamps. The compliance_status will be not_assessed until you run bias tests or generate a report.
3

Update model metadata

If your model advances through deployment stages or changes owner, update its record with mode: "update". Only the fields you supply are changed.
{
  "tool": "models",
  "arguments": {
    "mode": "update",
    "model_id": "mdl_01j9z...",
    "owner_name": "Model Risk Office",
    "version": "2.1.0"
  }
}
Updates emit a model.updated audit event automatically.

​
Required vs optional fields

FieldRequiredNotes
nameYesHuman-readable name for the model. Appears in report headers.
model_typeYesFor example: classification, regression, generative.
use_caseYesPlain-language description of what the model decides.
risk_tierYeshigh, limited, or minimal — maps to EU AI Act risk categories.
versionNoSemantic version string. Recommended for regulated deployments.
descriptionNoLonger narrative description. Used in model cards and reports.
owner_nameNoTeam or person accountable for the model.

​
How model IDs work

Every model registered in ComplyHat receives a globally unique ID in the form mdl_01j9z.... This ID is the primary key for all downstream operations:
  • Bias tests — pass model_id to bias_tests with mode: "run" to evaluate the model against a dataset.
  • Reports — pass model_id to reports with mode: "generate" to produce a framework-specific compliance report.
  • Drift monitors — pass model_id to drift with mode: "monitors_create" to start tracking distribution shift.
  • Model cards — pass model_id to model_cards with mode: "create" to generate a structured model card.
If you register the same logical model at multiple versions, each registration gets its own mdl_... ID. Keeping versions separate lets auditors compare bias test results and report histories across versions independently.
The compliance_status field on a model is updated automatically when you run bias tests (needs_review on pass, non_compliant on any failure). You do not need to update it manually.

​
List all registered models

To see every model in your organization’s registry, use mode: "list". 
{
  "tool": "models",
  "arguments": {
    "mode": "list"
  }
}
Next steps: Once your model is registered, generate a compliance report or head to the tool reference to see every mode available on the models tool.