Use cases

Perissos covers an AI agent from design to retirement. Not every stage is shipped. Each part below says how far it is built today, the stage table on our About page gives more detail for the stages it covers, and we would rather you check than take our word for it.

See the stage table on our About page.

Live you can do this on perissos.ai today, at the route named · Partly live some of it runs today, and the line says which part · Planned designed, not built

If you build or deploy AI agents

  1. Before you build

    "What will bind this agent?" Live

    Know which EU AI Act obligations apply before you write it, with the provision that makes each one bind.

    The free readiness check runs with no signup. The itemised list, each duty with its citation, opens on screen once you give a work email address and tick the consent box, and is emailed to you once you confirm that address.

    Where it runs today: /assessment

  2. Before it ships

    "Our customer wants to know what our agent can do." Partly live

    Publish a signed Agent Passport they can verify: what the agent is, what it may do, who approved it, and how it is stopped.

    An agent's record (what it does, its autonomy, its tools, its owner and who can halt it) can be published, and on paid plans included in a signed export that anyone holding it can verify offline against our published key. Who approved the agent is not recorded yet.

    Where it runs today: /app/agents (signed in), /app/trust (signed in), /app/connectors/export (signed in), /.well-known/export-signing-key

  3. While it runs

    "Our agent takes real actions." Planned

    Keep it inside its limits at the moment it acts, with each decision kept as dated evidence.

    Our own assistant asks before it acts, and a tool it claims without a live grant is refused before the action runs. That is enforcement of this product's assistant, not of the agents you deploy.

  4. Testing

    "Is our agent safe against misuse?" Planned

    Adversarial testing under your authorisation, with results sealed as dated evidence for your auditor.

    Consented testing is designed, not built.

  5. Audit

    "We need to be audit ready." Live

    Take ISO/IEC 27001 and ISO/IEC 42001 from gap to evidence to a readiness verdict that names what is missing.

    Signed in, you keep your management-system records and run the Gate, which gives each ISO/IEC 27001 and ISO/IEC 42001 requirement it covers its own verdict and says what is not ready. Most Annex A controls are met by attaching evidence files, which you can do today. We help you reach audit readiness; we do not certify or audit you.

    Where it runs today: /app/aims (signed in), /app/gate (signed in)

  6. When the law moves

    "Did the rules just change for us?" Partly live

    EU AI Act amendments and application dates tracked against your systems.

    The Act's application dates are published, cited. When you run the Gate, a duty that does not apply yet reads Not yet applicable, with the date it starts and its citation. Telling you when an amendment changes the rules for your systems is not built yet.

    Where it runs today: /eu-ai-act-dates, /app/gate (signed in)

  7. After it ships

    "Something went wrong." Partly live

    Post-market monitoring and serious-incident readiness that answer to the Act's own deadlines.

    On paid plans, an incident you record starts the reporting clocks it triggers under the rules we hold. Monitoring your agent after release is not built yet.

    Where it runs today: /app/incidents (signed in)

  8. Retirement

    "We are switching this agent off." Partly live

    Retire it with its records exported, retained or erased as the law requires.

    Decommissioning an agent is terminal and dated, and its record is kept. A workspace owner can export the whole workspace or delete the whole account. Exporting or erasing one agent's records, and retention periods set by law, are not built yet.

    Where it runs today: /app/agents (signed in), /app/account (signed in)

If you buy AI

  1. "Can we trust this vendor's AI?" Partly live

    A cited read of a vendor's AI from public evidence, with no score and the vendor's side shown.

    You can read a company's published AI surface with no account: each item observed, not observed or could not be read, with its source, and no score. Signed in, the read also shows the vendor's own trust profile, but only once the vendor has claimed its page and made its profile visible.

    Where it runs today: /vendors, /app/vendors (signed in)

  2. "How do our agents look to buyers?" Partly live

    See your agents from the outside, the way a customer's security team would, bound to the obligations each finding implicates.

    You can run the same public read on your own company that a buyer gets. Reading each agent, and tying each finding to the obligations it implicates, is planned.

    Where it runs today: /vendors