Skip to content

Status ledger

Verified. Tested.Not yet. Not done.

Every claim, with its evidence.

Message us

Verified

Run for real, with a real service or program. The evidence is kept.

  • Authenticator app sign-in

    Codes checked against an independent implementation.

  • Single sign-on

    OpenID Connect, with one identity provider. Others are documented, not tested.

  • Encrypted credentials

    Stored secrets encrypted with your key, with rotation.

  • Tamper-evident audit log

    Detects changes in a copy. It is not write-once storage.

  • Safe start in production

    Refuses to start with a missing key or an insecure address, naming the setting.

  • Locked-down container

    No root, read-only, no extra privileges. Scanned with no high or critical findings at the time.

  • Monitoring export

    Traces, metrics and logs received by a real OpenTelemetry collector.

  • Security log export

    Delivered as files. Ingestion by a real security system is not proven.

  • Connecting an agent platform

    One real run completed end to end. Failures and rate limits against the live service are not proven.

  • Agents over the A2A protocol

    Against the official reference agents only, not a live third-party agent.

  • Deployment on a virtual machine and Kubernetes

    In test environments. Nothing has run in a customer environment yet.

Tested

Covered by automated tests that use simulated agents and models.

  • Roles

    Administrator, operator and viewer, enforced on the server, with a test that fails if a route lacks a role.

  • Passkeys

    In a browser with a virtual authenticator. Hardware keys are not proven.

  • Login protection

    Throttling per email and per address, sessions that expire and can be revoked.

  • Approvals

    Written reason, approval rules by amount, risk, role and hour, two-person approval, assisted and manual takeover.

  • Emergency stop

    Suspends every approved agent at once and resumes only those it stopped.

  • Checking answers and reports

    Mark answers correct or incorrect with a reason. Accuracy computed from the checks.

  • Evaluations

    Fixed questions, scheduled runs, regression alerts. Quality with a live model is not proven.

  • Knowledge

    Versioned, published explicitly, searchable. A differently worded paraphrase is not found.

  • Issues

    Problems tracked from an answer or a run until fixed.

  • Register of AI systems

    Owner, purpose, model, data and a risk class chosen by the customer. Not a legal determination.

  • Personal data detection

    Emails, IBANs, tax numbers and card numbers. Names in free text are not found.

  • Retention, export and erasure

    Off until an administrator sets a policy. Nothing is deleted by default.

  • Cost and budgets per agent

    Alert or block at a monthly limit. Costs come from what the agent reports.

  • Notifications

    In-app, email, chat webhook and push. Email and push tested against local stand-ins only.

  • User provisioning

    SCIM request shapes for two identity providers. The conformance run is not clean and no real provider was connected.

  • Service tokens

    Scoped, expiring, revocable. The secret is shown once and stored hashed.

  • English and Portuguese

    Each person picks the language.

Not yet proven

Written or planned, but not run for real, or blocked by something outside our control.

  • Signed release images

    Signing was checked with a local key. The release pipeline has never run.

  • Real mail and push services

    Not run against a real mail provider or a browser vendor’s push service.

  • Evaluation quality with a live model

    The judge ran against the simulated model only.

  • A live third-party agent

    Blocked: no endpoint or token to test against.

  • Resuming a run that waits in an external platform

    Blocked: that platform’s public interface has no resume.

  • Cloud key services

    Not integrated. You provide the key.

  • Settings in the interface

    Some settings are still configuration values. Moving them into the interface is planned.

Not done

It does not exist. We say so.

  • Certifications

    No SOC 2 report, no ISO 27001 or ISO 42001 certificate.

  • External penetration test

    None has been done.

  • Legal assessment

    GDPR and EU AI Act obligations have not been assessed. Dates quoted by anyone need counsel’s confirmation.

  • Signed data processing agreement

    Draft templates exist. Counsel has reviewed none.

  • High availability

    One active instance per customer, with backup and restore.

  • Customer deployments

    No version has been released to a customer environment.

Spot a gap?Tell us.

Send us the hard question.