Ardoq’s AI Is Grounded In Our AI Principles

Enterprise architecture decisions routinely depend on 10, 20, even 50+ connected facts — application dependencies, risk ratings, ownership, compliance status, migration plans. With a standard LLM answering each reasoning step at around 92% accuracy, a question requiring 10 connected facts has only a 43% chance of being correct. That's a coin flip. Ardoq's neuro-symbolic approach fixes this: AI generates the reasoning logic, and the graph executes it deterministically. The result is accurate, consistent, and auditable, not confident-sounding guesses.

Most EA tools add AI on top of flat metadata, static diagrams, or disconnected document stores. Ardoq's AI is built on a graph-native architecture that encodes real enterprise relationships, enforces schema constraints, and via GraphLake, preserves temporal context and decision traces. That means AI outputs in Ardoq are traceable to specific, structured data points in your live architecture. Other tools give you fast-sounding summaries. Ardoq gives you decision-grade answers you can act on and audit.

It means AI in Ardoq acts within the boundaries your architecture defines, not outside them. In practice: permissions are enforced before every model call, so AI can only access what a user is authorised to see. AI-generated outputs land in draft or scenario state by default, requiring human review before any change applies to the live architecture. And as Ardoq introduces higher levels of autonomy over time (like agents that can execute tasks) governance scales with it. Autonomy and governance aren't in tension; they're designed together.

Most architecture tools only show you the current state. But enterprise decisions are made in time and the questions that matter most are often retrospective: what did our architecture look like six months ago? Why did we decommission that application? What changed between this quarter's audit and last year's? Ardoq's temporal intelligence, powered by GraphLake, lets you rewind the architecture graph to any point in time, surface the reasoning behind decisions, and trace how things changed. For regulatory purposes (EU AI Act, SOX, ISO compliance) this audit trail is structural, not a manual afterthought.

No. Our differentiation is not in training frontier models, it's in what we do with them. Ardoq uses a model-agnostic architecture that routes queries to the best available LLM based on task type, cost, latency, and data privacy requirements. Our IP is in the enterprise ontology, the knowledge graph structure, the neuro-symbolic reasoning layer, and the governance enforcement that sits around every model call. This also means customers aren't locked into any single AI vendor's roadmap, as the model landscape evolves, Ardoq moves with it.

MCP (Model Context Protocol) is an open standard for connecting AI tools to external data sources. Ardoq was the first EA vendor to GA an MCP Server and has seen exponential usage growth in 2026. In practical terms: the Ardoq AI Gateway exposes your architecture data to external AI tools (Microsoft Copilot, Gemini, ChatGPT, custom enterprise agents) in a structured, permission-aware way. It's the Graph-Grounded and Governed Autonomy principles made real for any AI tool your organisation uses. External AI gets genuine architectural context, filtered by your role-based permissions, not a raw text dump of your documentation.

Yes and several are already live. The Foundation Insights Agent, Data Ingestion Agent, and a suite of Solution Agents are in active use across the customer base. The Spring 2026 roadmap brings Custom Agents to GA letting teams build their own scoped agents on top of Ardoq's architecture data along with the App Rationalization Agent and Disaster Recovery Blast Radius Agent. All agents operate within the Governed Autonomy principle: they act within defined permissions, produce outputs in draft state, and require human approval before changes apply to the live architecture. Autonomy increases incrementally, Ardoq's position is that agentic AI is only safe when governance is built into the architecture, not bolted on later.