Skip to content
Guides & References

Rolling out Iris to a network team

A phased plan for introducing Iris and MCP agents to a network team: pilot, provider choice, knowledge entries, measuring use, and widening access.

Iris: AI CenterUpdated
On this page

Rolling out Iris to a network team

Turning Iris on is a five-minute job. Getting a team to trust it and use it well takes a few weeks. A phased rollout answers the security team’s questions first, gives engineers a feature that helps them on day one, and builds the team knowledge that makes answers fit your network.

Phase 1: decide (week 1)

  • Provider. Agree where configs may go: on site, your cloud tenant, or a public API. See Choosing a provider.
  • Features. Decide which Iris features to enable. Compliance narration sends no config text, so it’s often approved first.
  • Agents. Decide whether MCP agents are in scope yet, and which clients and model providers are acceptable.
  • Security review. Walk through the security review with your security team.

Phase 2: pilot (weeks 2 to 4)

  • Small group. Three to five engineers who handle configs daily. Give their roles aicenter.view; each accepts the terms on first use. Add aicenter.mcptoken for those who’ll connect an agent, so they can create their own MCP tokens.
  • Start with two features. Config analysis and drift narration help on day one. Turn the others off until the pilot asks for them.
  • Write the first knowledge entries. Naming standard, two or three conventions, one golden config. See The context harness.
  • Collect bad answers. Ask the pilot to note answers that were wrong or generic. Most are fixed by a better question, a knowledge entry, or a more capable model.

Phase 3: measure

The AI audit trail records every request. There’s no screen for it in this release; an administrator can summarise it from the database:

SELECT capability, COUNT(*) AS requests, COUNT(DISTINCT user_id) AS users,
       SUM(prompt_tokens) AS prompt_tokens, SUM(completion_tokens) AS completion_tokens
FROM ai_audit_logs
WHERE created_at >= NOW() - INTERVAL '30 days'
GROUP BY capability;

The syntax shown is for PostgreSQL; on MySQL use INTERVAL 30 DAY. Which features get used, by how many people, tells you what to keep and what to explain better.

Phase 4: widen

Frequently asked questions

How should we start using rConfig's AI features?

With a small pilot group, one provider your security team has approved, and the features they'll use most, usually config analysis and drift narration. Add knowledge entries for your conventions, then widen access once the pilot's questions are answered.

How do we know if people are using Iris?

The AI audit trail records every request with user, feature, model and token counts. Counting requests per feature and per user over the pilot shows what's used.

When should we introduce MCP agents?

After Iris's built-in features are bedded in, and once you've decided which agents and model providers are acceptable. Start with engineers who already use an AI coding agent.

Type to search.