Agent Registration Step by Step

The agent registration wizard is the compliance officer's tool of choice. One page, seven sections that adapt to your plan — from 2 sections on Community (~30 seconds) to all 7 on Enterprise (~5-10 minutes).

Start the wizard

1. Navigate to Agent Registry in the sidebar.
2. Click New Agent (top right).
3. The registration page opens — settings-style layout with explanations on the left and inputs on the right.

Alternatively: Pick a template from the gallery to start with sensible defaults (Customer Support, HR Screening, Code Assistant, etc.).

Section 1 — Identity & Templates (all plans)

Templates: Picking a template auto-fills description, data-protection profile, and other defaults. You can tweak everything afterward.

Section 2 — AI-Assisted (Pro+)

The wizard includes an AI-assist step that fills in tedious governance fields based on the agent name and description.

• Click the ✨ sparkles button next to Description to generate a description from the agent name.
• Click ✨ AI-assist this section at the bottom of each section to pre-fill fields where the AI is confident — purpose statement, suggested risk level, suggested data-protection level.

The AI suggestions are assistance only — every field is editable, and nothing is saved until you click Register agent. The AI-assist endpoint is the same one that powers the NL Policy Builder.

Section 3 — Protection Profile / Data Protection

Community

Pick a protection profile from three cards:

Pro and above

Three data-protection levels replace the profile:

Section 4 — Responsibility Chain (Business+)

Three accountable people per agent:

Dropdowns show every team member with editor or higher. For HIGH-risk agents, all three fields are mandatory.

Section 5 — Provider / Deployer Role (Business+, EU only)

The EU AI Act distinguishes providers (who develop / market the AI system) and deployers (who use it in their organization). Different obligations apply to each.

The wizard pre-selects Deployer for templates that use external APIs (OpenAI, Anthropic, Google), and Provider for templates that wrap your own models. Override as needed — your legal team should weigh in for HIGH-risk agents.

Section 6 — Risk Classification (Business+, EU only)

Pick the risk level from four cards:

For HIGH: Pick one of the 8 Annex III domains (Biometrics, Critical Infrastructure, Education, Employment, Public Services, Law Enforcement, Migration, Justice). Human oversight is enabled automatically and cannot be disabled.

Section 7 — Compliance Screening (Enterprise, EU + multi-framework)

The screening section runs the agent against multiple regulatory frameworks. You answer a small set of yes/no questions per framework; the wizard surfaces which obligations are triggered.

EU AI Act Art. 5 (prohibited practices)

1. Does the agent use social scoring or evaluate persons based on their social behavior?
2. Does the agent use subliminal manipulation to influence decisions?
3. Does the agent use real-time biometric identification in publicly accessible spaces?
4. Does the agent use emotion recognition in workplaces or educational institutions?

Any Yes → registration blocked. The wizard surfaces the specific Art. 5 paragraph and recommends consulting Legal.

EU AI Act Art. 6 / Annex III (high-risk classification)

The wizard walks through the 8 Annex III domains and asks targeted yes/no questions per domain. A Yes in any domain auto-classifies the agent as HIGH risk (overriding Section 6 if necessary) and unlocks the FRIA workflow.

Colorado AI Act (US — when applicable)

Active when Provider/Deployer role indicates US operation. Five yes/no questions covering:

• Consumer-facing consequential decisions
• Personalized pricing or eligibility
• Algorithmic discrimination risk
• Impact-assessment requirements

A Yes triggers Colorado AI Act controls — algorithmic-impact-assessment workflow, disclosure obligations, and biennial bias audit reminders.

Singapore PDPA (APAC — when applicable)

Active when the deployment jurisdiction includes Singapore. Covers AI-specific PDPA obligations: explanation requirements, opt-out mechanisms, and DPO-attestation requirements.

Outcome

The screening result lands as a screening_result JSONB field on the agent record:

{
  "eu_ai_act": { "risk_level": "HIGH", "annex_iii_domain": "employment", "fria_required": true },
  "colorado_ai": { "applies": true, "biennial_audit_due": "2027-12-01" },
  "pdpa_sg": { "applies": false }
}

The compliance dashboard reads from this field to compute the readiness scores per framework.

Section 8 — Success & Developer Handoff

After successful registration:

1. Agent API key — visible on the Integration tab with a freshKey banner on first registration.
2. Setup instructions — framework-specific integration guides (LangChain, CrewAI, OpenAI Agents, etc.).
3. Test command — a copy-paste curl to confirm connectivity.

Open the agent detail and share the setup instructions URL with the developer.

Plan matrix

Three layers of defaults

The wizard pulls from three sources, lower beats higher:

1. Project defaults — Admin configures once under Settings → Agent defaults.
2. Template defaults — Template overrides project defaults for the fields the template owns.
3. Per-agent adjustment — User overrides the rest.

The fifth agent should only need a name and description — everything else comes from defaults.

Imported agents (Cross-Platform)

Agents imported via Cross-Platform Import (Azure AI Foundry or Google Vertex AI) arrive with identity fields pre-filled (name, model, tools, instructions). The governance sections still need to be completed — open the agent detail and run the wizard from there.