Thalus Logo
Product updates

Thalus AIR is now a connector in Claude

Create projects, add evidence, run EU AI Act risk assessments and read your portfolio, all from a Claude chat.

· 2 min read

Claude connector

Thalus AIR

Open in Claude

Know whether the AI you’re building is high-risk, before it ships.

  • Classify a system under the EU AI Act and see exactly which criteria set the tier.
  • Upload project briefs, model reports and technical documents as evidence, then assess against them.
  • Get the controls each risk needs, with EU AI Act, NIST AI RMF and ISO 42001 references.
  • Answer the open questions the tier depends on in the chat. The same assessment updates, with no new credit.
  • Find projects across your portfolio with no completed assessment.
  • Pull a full report or executive summary straight into whatever you’re drafting.
  • Every result traces from evidence to facts to rules to outcome, so you can show an auditor, a customer or your board why a decision was made.

New to AIR? Create an account at air.thalus.ai. Your first assessment is free.

What's new

Thalus AIR is now listed in the Claude connectors directory. Connect it once and Claude works with your AIR workspace directly: the same domains, projects, documents and assessments you see at air.thalus.ai, now in the conversation.

So the owner of an AI system can check whether it is high-risk while working in Claude, before the build is finished. To see a full assessment run in one Claude chat, watch the product tour.

What you can do from a chat

  • Set up a project and add its documents. Project briefs, intake forms, feasibility studies and model reports, as Word, PDF and other common files. AIR reads each one so every finding can cite it.
  • Run a risk assessment and follow it to the end. AIR finds the risks, links the controls each one needs and sets the EU AI Act tier, with the evidence behind every finding.
  • Close the gaps it cannot decide alone. When the tier depends on facts your documents do not state, AIR lists them as open questions. Answer them in the chat and the tier is worked out again on the same assessment, with no new credit.
  • Read the whole portfolio. Ask for a domain's risk states, its mix of tiers, the risks that come up most and the controls suggested most, and record decisions on a project in your own words.

Five things to ask first

Plain language works. These are the prompts we use in demos:

  1. "List my AIR domains and show which projects still need their first assessment."
  2. "Create a project called Clinical triage pilot and upload these documents."
  3. "Run an EU AI Act risk assessment on them and tell me when it is done."
  4. "Which open questions are blocking the tier? Answer what you can from the documents and show me the rest."
  5. "Give me the top five risks and the first controls to put in place, with the evidence for each."

How to add it

  1. Open the AIR listing at claude.ai/directory/air.
  2. Connect it, and sign in with your AIR account or create one.
  3. Turn AIR on in a chat and ask your first question.

Good to know

  • Each assessment uses one credit, and your first one is free. Answering open questions afterwards updates the same assessment and uses no credit.
  • AIR's risk themes are mapped to the MIT AI Risk Repository, and its controls carry EU AI Act, NIST AI RMF and ISO 42001 references.
  • You review every finding before you act on it or share it.
  • Working in a code editor instead? See Connect AIR to Claude and other MCP clients.

Read next