Skip to main content

Identigy Access Copilot

The governed AI layer for your IGA

Incident investigation, audit evidence and diagnostics in minutes — as the signed-in user, under native authorization. Bring your own LLM over the open MCP standard. Chat, Claude Desktop, Cursor or your ITSM — the same governed access.

Talk to an architect

What the governed AI layer for IGA does

  • Governed AI access to your IGA

    A layer through which any AI reaches your identity data — strictly within the permissions of the person asking. Secrets are masked, every request lands in the audit trail. Not a standalone chatbot.

  • MCP server: 11 tools, read-only

    An operation catalog over the open Model Context Protocol: search and read objects, object schema, audit log, connector tests, log tailing with secret masking. Not a single write operation.

  • Investigation, audit and diagnostics in plain language

    “Who has access to system X”, “what changed on this account in the last 24h”, “why is LDAP not responding” — answered against live system data in seconds, without hand-building filters.

  • RAG over docs and policies

    Semantic search across platform documentation and the customer's own runbooks and policies. Answers cite their sources: asked “how does this work”, the Copilot replies with a citation, not a guess.

  • The LLM of your choice

    Works with the model your organization has adopted — on-prem/self-hosted or cloud, your call. On-prem keeps queries and system data inside the perimeter for data-residency cases.

  • Four ways to connect

    The built-in chat in the IDM UI is only one entry point. The same operations are available to external AI clients (Claude Desktop, Cursor, IntelliJ), to your ITSM chat, and to any future MCP client.

Access Copilot — chat

read-only

Who in Department A currently has access to the Billing system?

Right now 3 people in Department A have access to Billing:

j.smith  · J. Smith  · Billing operator

a.novak  · A. Novak  · Billing operator

d.kim    · D. Kim    · Billing admin

Tools: object schema → object search (roles and department)

And who had access on Jul 7, 2025?

Per the audit log, on Jul 7, 2025 access was held by the same three plus M. Riley (m.riley) — revoked Sep 2, 2025 on transfer out of Department A.

Tools: audit log search → event with change deltas

as: auditor · no privilege escalation

Dialogue simulated from the product's demo scenarios; the data is illustrative.

How it works

How the AI layer works with your IGA

An administrator or auditor states the task in plain language. The model calls the Copilot's tools, pulls data from the system — as the signed-in user — and builds a complete answer grounded in facts and documentation.

Access Copilot architecture: the governed AI layer Four AI clients — the chat in the IDM UI, external AI clients, the service-desk chat and any future MCP client — connect over the MCP standard to the governed Access Copilot layer. The layer acts as the signed-in user and reaches the IGA system and the knowledge base. All 11 operations are read-only; with an on-prem LLM data never leaves the customer perimeter. ANY AI — ONE LAYER Chat in the UI a page inside the IDM AI clients Claude Desktop · Cursor Service-desk chat ITSM / Help Desk Future client any MCP MCP — the open connection standard ACCESS COPILOT — THE GOVERNED LAYER User's permissions Secret masking Every request audited 11 tools — read-only Agent rate-limiting as the signed-in user citations, not guesses Your IGA objects · roles · audit log · logs midPoint · SailPoint · Oracle · … Knowledge base (RAG) platform docs · your runbooks answers cite their sources CUSTOMER PERIMETER with an on-prem LLM, data never leaves the perimeter — for residency-bound environments
The built-in chat is just one of four entry points. One layer: the same permissions, masking and audit trail for every AI client.
  1. The user states the task in plain language
  2. The Copilot calls the right MCP tools in your IDM
  3. The system returns data (objects, audit, logs) — strictly within the user's permissions
  4. The Copilot assembles a complete answer, grounded in facts and docs

The result: routine access, audit and diagnostics questions are answered in seconds — without manual filters or hand-offs between teams. Heavy filtering runs as a single database-side query, not an AI-side loop over records.

The Copilot itself is a working case of AI agent identity done right: no standing credentials, permissions scoped to whoever asked the question, every call in the audit trail.

The catalog: 11 tools

read-only

Search & read

  • Object types
  • Search by name and attributes
  • Object schema
  • Full object card

Audit

  • Audit log search
  • Event with change deltas

Diagnostics

  • Resource connection test
  • Log tailing

Knowledge (RAG)

  • Search over docs and policies

Plus two service tools — a connection check and server date/time: eleven in total, all read-only. Search results come back as a compact summary (name, type, status), with the full object card a separate call: the conversation history never balloons by tens of kilobytes. The full tool specification is available in the technical documentation on request.

Integration

Works with your IGA — no platform replacement

Shipped as a separate AI layer, the Copilot plugs into your existing identity system over the open MCP standard and connectors — no platform replacement.

Evolveum midPoint

Built

Our working product Midpoint AI is a midPoint overlay — the whole AI layer is proven here.

SailPoint (IIQ / ISC)

Per project

Through the SailPoint MCP server (GA 2025) — for your current and legacy installations.

Oracle (OIM / OIG)

Per project

Through connectors / MCP — including as an AI overlay on a platform you plan to migrate off.

For on-prem IGA the AI layer and data stay inside the perimeter — even when the platform itself is legacy. A strong fit for regulated, residency-bound environments.

Built, not slideware

A working reference, then your platform

The whole AI layer is proven in our working product Midpoint AI — an overlay for Evolveum midPoint. What you see in the Copilot is built and tested on a real IGA system, then delivered on top of your platform.

See our midPoint practice →

Where the market is

SailPoint shipped its own MCP server in 2025 — the direction is validated by the industry. Our layer goes wider: 11 read-only tools spanning operations, audit and diagnostics — against 4 access-request tools in SailPoint's official server — and an on-prem model when residency demands it.

Connect Claude Desktop, Cursor or any MCP client — the reasoning lives in the client, the governed data access lives in the Copilot.

The platform is investing in AI too

Evolveum midPoint is investing in AI

The open-source platform our reference implementation is built on isn't standing still. Evolveum — the company behind midPoint — ran an EU-funded AI programme, so our Copilot sits on a foundation with real, funded AI momentum.

  • midPilot — an EU-funded project (2025–2026, Slovakia's Recovery & Resilience Plan, NextGenerationEU): AI for application onboarding — connector generation, attribute mapping and correlation.
  • Finalist, AI Awards 2026 — Trustworthy AI category.
  • Gartner IAM Summit, London (March 2026) — Evolveum as a Silver Sponsor.
  • A 153-page architecture & research report (Milestone 1) — public and detailed.

The two AI layers are complementary. midPilot speeds up onboarding a new application (at configuration time); Access Copilot runs on top for day-to-day operations — natural-language queries, MCP, RAG and audit. One platform, two layers of AI.

Read Evolveum's midPilot research report (PDF) ↗
Security and residency

What will the AI see? Exactly what the user sees

The security team's first question is answered by architecture, not configuration: the layer runs under the IDM's native role model, with no privileges of its own.

The AI acts as the authenticated user: it sees what they see, with no privilege-escalation path by design
All 11 operations are read-only. The Copilot cannot create or modify records; write operations can be enabled for your project as a separate decision
Use the LLM you've standardized on — on-prem keeps queries and system data inside your network
Secrets (passwords, keys, tokens) are masked before they reach the answer; confidential fields are stripped from audit records
Sensitive operations (e.g. reading logs) sit behind a separate permission
Rate-limiting guards against runaway agents — request throttling on the MCP endpoint and a minimum interval between chat requests

User: auditor

  • Audit log
  • User and role cards
  • System logs — separate permission

AI acting as the auditor

  • Audit log
  • User and role cards
  • System logs — separate permission

The AI's permissions = the user's permissions. Always. There is no "AI account" with elevated access.

A separate guardrail against hallucinations: when the documentation has no answer, the Copilot says “not found” instead of a plausible guess. Answer generation can be stopped at any moment.

Scenarios

One layer — six roles

From employees to engineers: everyone asks in plain language and gets least-privilege answers — exactly what their permissions allow. The roles and restrictions are the ones already configured in your IGA.

Employee

Find the right role to request and track what's happening with your request — without a ticket to the help desk.

  • “Which role gives access to the CRM — and which roles do I already have?”
  • “What's the status of the approval for my Billing access?”

Line manager

Your team's access at a glance and quick pulls before a review — within the permissions a manager already has.

  • “Who on my team has access to Billing, and through which roles?”
  • “Summarize my direct reports' access to the finance systems.”

Helpdesk L1–L3

Fewer escalations: incident triage with advice from the vendor docs — resolved on the first line, not three days later on the third.

  • “Why wasn't an account created for the new hire? Check the audit log.”
  • “What does this connector error mean and how is it fixed? Cite the source.”

Security analyst & auditor

The average breach lifecycle is 241 days (IBM, 2025). Security events and audit evidence in minutes — with citations.

  • “Who changed user passwords this week? Initiator → target → when.”
  • “Find superusers whose password hasn’t changed in over 90 days.”

IGA administrator

Inventory, diagnostics and “where do I configure this” guidance — without hand-navigating the role tree or the docs.

  • “Check that all resources are reachable. If one isn’t, explain why.”
  • “Where do I configure auto-disable for leavers? Steps and links, please.”

Developer / engineer

Object schema, live configuration and midpoint-samples examples — right inside Cursor, Claude Desktop or IntelliJ.

  • “Show the UserType field schema and a sample resource XML configuration.”
  • “Debug this sync error from the logs and find a similar sample config.”

Talk about a governed AI layer for your IGA

We'll demo it on a real system and come back within 48 hours with a fixed-fee proof-of-value scoped to your platform.

Talk to an engineer →
FAQ

Frequently asked questions

What is Identigy Access Copilot?

A governed AI layer on top of your identity system (IGA). It gives any AI — the built-in chat, an external AI client, your service-desk chat — access to identity data strictly within the permissions of the person asking, with secret masking and a full audit trail of every request.

Is it a chatbot?

No. The built-in chat is just one of four ways into the layer. The same 11 tools are available to external AI clients (Claude Desktop, Cursor, IntelliJ) and to your ITSM chat over the open MCP standard.

Can the Copilot change data in the IDM?

No. All 11 tools are read-only: the Copilot cannot create or modify records. That is how we build trust at the first stage; write operations can be added for your project without changing the architecture.

Which IGA platforms does it work with?

Natively with Evolveum midPoint. With SailPoint and Oracle via the open MCP standard and connectors, per project. No platform replacement — it is an AI layer on top of what you already run.

Which LLM does the Copilot use?

The model your organization has adopted — on-prem/self-hosted or cloud, your choice. On-prem keeps queries and data inside the perimeter for data-residency cases.

What is MCP for identity governance?

Model Context Protocol — an open standard to connect external AI clients to your IDM. The Copilot exposes 11 read-only tools: search/read objects, object schema, audit log, resource tests, RAG over docs, log tailing.

Is it secure to give AI access to the IDM?

The AI sees exactly what the signed-in user sees: no privilege escalation, every operation read-only, sensitive actions behind a separate permission, secrets masked before they reach the model, request rates limited.