ComplyHat: Docs

Eleven entity tools plus guidance. Each entity tool dispatches by mode. All callable via https://complyhat.ai/api/mcp. Audit-tagged citations ([EXTRACTED], [INFERRED], [AMBIGUOUS]) appear in every prose field.

guidance , the meta-tool

Returns the canonical SKILL.md body for any skill or agent. Start here when your host wants to know what tools to call for a given compliance flow.

Argument	Type	Description
`kind`	`"skills"` or `"agents"`	Resource namespace.
`slug`	string	Skill or agent slug.

{
  "tool": "guidance",
  "arguments": { "kind": "skills", "slug": "bias-test-prep" }
}

The same content is exposed as MCP resources at complyhat://{kind}/{slug} for hosts that auto-include resources.

Skills
Agents

Slug	Purpose
`bias-test-prep`	Dataset preparation for bias testing.
`compliance-checklist`	Per-framework checklists.
`compliance-wiki-ingest`	Fold a local document (PDF, DOCX, policy, regulator letter) into compliance memory.
`model-documentation`	Model card authoring.
`regulatory-mapper`	Map use cases to frameworks.
`risk-classification`	EU AI Act risk-tier classification.

Slug	Purpose
`bias-tester`	Runs bias tests and surfaces results for host review.
`data-governance-auditor`	Audits sources and lineage.
`drift-detector`	Submits drift snapshots and surfaces drift scores.
`explainability-tester`	Runs explainability evaluations.
`model-card-writer`	Drafts and updates model cards.
`report-generator`	Drafts framework reports for host composition.

Entity tools

Eleven bundles. Each dispatches by mode. Click a row to expand.

models , register and inspect AI models

The model registry. Root entity most other tools reference.

Mode	Description
`list` / `get`	List or retrieve registered models.
`register` / `update`	Register a new model or update metadata.

model_cards , structured, audit-tagged model documentation

Mode	Description
`list` / `get`	List or retrieve model cards.
`create` / `update`	Create from a model registration, or update.

bias_tests , fairness tests against registered models

Disparate impact, statistical parity, equal opportunity, predictive parity. See methodology for the underlying statistical methods.

Mode	Description
`list` / `get`	List tests for a model or get one result.
`run`	Run a bias test against a model and dataset for one framework. Pass the framework per call; ComplyHat applies that framework’s thresholds.

drift , monitors, snapshots, alerts

PSI, KS, and chi-squared two-sample drift.

Mode	Description
`monitors_list` / `monitors_create`	List or create drift monitors.
`snapshots_list` / `snapshots_submit`	List submitted snapshots, or submit a new production snapshot from the host.
`alerts_list_by_model`	Drift alerts for a model.

explainability , LIME and coalition-attribution runs

Mode	Description
`list_by_model` / `get`	List runs for a model or get one’s results.
`run`	Trigger a new run.

adversarial , robustness findings (host-submitted)

ComplyHat does not generate attacks; it stores host-submitted findings with audit-tagged citations.

Mode	Description
`list_by_model`	List adversarial findings for a model.
`get`	Get one finding by ID.
`submit`	Submit a new finding.

reports , compose, finalize, and download compliance reports

{
  "tool": "reports",
  "arguments": {
    "mode": "start_draft",
    "model_ids": ["00000000-0000-0000-0000-000000000000"],
    "template_slug": "eu_ai_act",
    "name": "Q2 EU AI Act validation"
  }
}

Mode	Description
`list` / `get`	List or retrieve reports (get returns approved_content once finalized).
`get_docx`	Render the finalized report as a DOCX from its approved_content.
`start_draft`	Begin a new draft: returns template + structured evidence + your compliance memory. Host composes prose.
`finalize`	Persist user-approved prose with a sha256 hash. Auto-appends an approved-positions block to your compliance memory.

See the report lifecycle page for the host-composes, user-approves flow. Signing and regulator submission happen in the customer’s GRC workflow outside ComplyHat.

frameworks , regulatory framework status, relevance, freshness

Mode	Description
`status`	Overall coverage status for the organization.
`check`	Check whether a framework applies to a model or use case.
`relevance`	Triage a pasted-in regulation text: score which tracked frameworks it touches and how it maps to your models.
`check_freshness`	Curated trusted regulator URL plus a `host_instruction` directive so the host agent verifies amendments at the source before relying on a stamped framework version.

audit_events , append-only audit trail

Mode	Description
`list` / `get`	List events with filters or get one by ID.
`emit`	Emit a new audit event from the host agent.

data_governance , sources and lineage

Mode	Description
`sources_list` / `sources_create`	List or register data sources.
`lineage_list` / `lineage_create`	List or create lineage entries.

wiki , your compliance memory (one master markdown doc)

One master markdown document per tenant: your accumulated compliance posture, returned on every reports.start_draft so new drafts stay consistent. ComplyHat stores markdown only and never calls an LLM.

Mode	Description
`read`	Return the whole document plus its version.
`write`	Replace the whole document. Read it, edit it, write it back. Pass `prev_version` for optimistic concurrency (a mismatch returns 409).
`append`	Add a markdown block to the end without re-sending the whole file.
`delete`	Clear the whole document, keeping its version history. For erasing your accumulated posture.

Documentation Index

​guidance , the meta-tool

​Entity tools

guidance , the meta-tool

Entity tools