2026-10-18
Added
- Decide which models and MCP servers your organization may use, find AI in use that nobody registered with Discover, and give each person one of five roles: Owner, Admin, Governance Lead, Member or Auditor.
- Send your applications’ and coding agents’ model traffic through one address that speaks the OpenAI, Anthropic and Gemini formats, with guardrails on every request and reply, budgets, rate limits, fallbacks, saved prompts, a semantic cache, MCP tools and request logs.
- The docs have an API reference for the AI Gateway, with every endpoint, header and error.
- Attaching a key to an AI system that is not live now warns you first: the key’s requests are refused until the AI system goes live.
- Keep a register of your AI systems, each with an owner, a risk rating worked out from seven questions and its EU AI Act category. Assess each one in a conversation with Altrum’s assistant, turn the risks it finds into guardrails, and approve it before it goes live.
- Follow cost, guardrail violations and tool use in Reporting, get alerts in the product and by email, and handle incidents, including serious incidents under the EU AI Act. Show auditors each AI system’s posture, quarterly access reviews, red-team runs, compliance reports and the audit trail.
- The account menu shows which version of Altrum you are using, so you can quote it to support.