Skip to main content

Deploy Your First Agent

This walkthrough uses the Ariftly dashboard to deploy the AI Readiness Agent and run your first task. From zero to a completed compliance audit in under 10 minutes.

Prerequisites​

Before you start:

  • An Ariftly account — sign up at app.ariftly.io
  • A GitHub account with at least one repository containing your AI-related code or documentation
  • (Optional) An AI provider API key — Anthropic or OpenAI — if you want to bring your own instead of using the shared compute tier

The AI Readiness Agent works best when it has access to at least one of: your codebase, model cards, data pipeline documentation, or existing compliance documents. If you have none of these yet, you can still run a gap analysis against a bare knowledge base to see what you need to produce.

Step 1: Open the Dashboard​

Go to app.ariftly.io and sign in.

If this is your first login, you will be prompted to name your organization and select your industry. This context helps the AI Readiness Agent understand which regulatory frameworks apply to your use case (EU AI Act, NIST AI RMF, etc.).

Step 2: Connect GitHub​

Navigate to Integrations → GitHub → Connect and click Install GitHub App.

You will be redirected to GitHub, where you can choose to grant Ariftly access to:

  • All repositories — recommended if you want the widest scan coverage
  • Select repositories — recommended if you want to limit scope to AI-specific repos

Grant access to the repositories most relevant to your AI stack — model training code, inference services, data pipelines, and any compliance documentation stored in version control.

What the agent reads

The AI Readiness Agent reads file contents to understand your codebase structure, identify AI libraries in use, detect hardcoded credentials or unsafe AI practices, and ground its questionnaire answers in real technical evidence. It does not write to your repositories or open pull requests without your explicit approval.

Step 3: Go to Agents → Deploy​

Click Agents in the sidebar, then Deploy New Agent.

Select AI Readiness Agent from the catalog.

Give it a name that reflects its scope — for example:

  • "Production AI Audit" — for scanning your production AI services
  • "Pre-deal Compliance Check" — for answering a specific procurement questionnaire
  • "EU AI Act Readiness 2026" — for a regulatory compliance initiative

Click Deploy. The agent appears in your agent list within a few seconds. Initial setup (indexing your connected GitHub repositories) runs in the background and typically takes 1–3 minutes.

Step 4: Run a Task​

From your agent's page, click Run Task → Full Audit.

The agent will:

  1. Pull the latest state of your connected GitHub repositories
  2. Scan model configuration files, inference code, and data pipeline definitions
  3. Cross-reference your codebase against EU AI Act requirements and NIST AI RMF controls
  4. Identify gaps in your documentation (missing model cards, no AI incident response policy, etc.)
  5. Score your readiness across six domains: risk classification, human oversight, data governance, transparency, security, and technical documentation
  6. Generate a draft procurement questionnaire response using the evidence it found

The full audit typically completes in 2–5 minutes, depending on the size of your connected repositories.

You can track progress in real time in Dashboard → Tasks. The status moves through pending → running → complete.

Step 5: Review the Artifact​

When the task completes, you will see two artifacts in the task panel:

ai_readiness.audit_report — the full compliance report, including:

  • An overall readiness score (0–100)
  • A breakdown across each domain
  • A list of specific gaps with severity ratings (Critical, High, Medium, Low)
  • Evidence citations — the exact files or sections your codebase is missing
  • Recommended remediation actions for each gap

ai_readiness.questionnaire_response — draft answers to the 15 most common procurement questionnaire questions, grounded in the evidence the agent found in your repos

Click either artifact to open it in the full viewer. You can copy sections, download as PDF, or share via a secure link.

Step 6: Approve or Modify​

If you triggered the task with an actual procurement questionnaire attached, the agent will also submit the response for your approval before anything is sent.

The Approvals inbox shows:

  • The questionnaire response draft
  • Which specific claims are backed by evidence from your codebase (shown as citations)
  • Which claims are marked as low-confidence (flagged in yellow — these should be reviewed carefully)

You can edit any answer directly in the approval view. Once you approve, the response is finalized and available to send to the procurement team.

If you ran a standalone audit without a specific questionnaire, no approval is needed — the report is available immediately in your artifacts panel.

What to do with the results​

A typical first audit surfaces 10–20 gaps. Don't be discouraged — most companies find the same categories of gaps:

  1. Missing model cards — no formal documentation of training data, intended use, or risk classification for AI-powered features
  2. No human oversight logging — AI decisions affecting users lack an audit trail
  3. Undocumented training data sources — the agent can't find DPAs or data lineage documentation
  4. Hardcoded or unrotated AI credentials — API keys in source code or stale access tokens

The audit report prioritizes these by business impact, not just severity. Gaps that would cause a procurement questionnaire to fail are flagged separately from general hygiene improvements.

What's next​