Reconstruct what your AI actually did.
When an AI action is questioned, rebuild it step by step: who acted, who approved, what changed. With proof.
Illustrative example
One AI action goes wrong.
Watch what it costs you.
The damage is rarely the action itself. It is your best people rebuilding the story from six systems, and the decision you still can’t defend at the end.
The case: a customer received a quote for €1,000. The quote approved internally was €10,000. An AI agent sent it.
Illustrative example. Your own numbers come from your pilot.
Proof is not the brake.
It is how AI gets the keys.
Most AI stays stuck in pilots because nobody can show it behaves. When every action leaves evidence, you can hand an agent the next level of trust, and take it back the moment the record says so.
- Act within boundsIt acts alone inside limits, and every action is on record.Not yet
- Act with approvalIt acts after a named person approves the exact action.Not yet
- SuggestIt drafts, a person decides every time.Unlocked
- ObserveThe agent suggests nothing yet. You watch what it would do.Unlocked
The aim: fewer failed runs, less wasted compute, approvals only where they matter. Illustrative.
If AI can move, it needs a Passport.
Three steps. No rip-and-replace.
- 1 · Where your agents already runConnect
Install the free app for your own AI agents, on your own device, or connect read-only to the systems a pilot agrees on.
- 2 · Every action, as metadataRecord
Which agent acted, what it did, who said yes. Never prompts, files or code.
- 3 · One record per actionProve
Rebuild an action step by step, each claim marked Proven, Reported, Disputed or Missing.
The same record. A different win for every desk.
Pick the role closest to yours and see what changes on your side of the table.
“Was this an attack, a bug, or an agent doing what it was allowed to do?”
Which identity acted, under whose authority, and where does the trail break?
Illustrative. Views across teams and shared exports need the Enterprise edition, which is coming soon.
Metadata first. Your secrets stay yours.
Never the content: no prompts, files or code. You hold the keys: you decide what leaves your environment. Honest by default: proven, reported, disputed or missing; evidence, not a legal certificate.
Programmes, ecosystems and contributions · participation, not endorsement
- Campus FoundersAI Start · Incubator · Resident · Accelerator
- ★ WLOUNGE Tech Awards 2025Top 3 finalist (Corporates Open Innovation, Berlin)
- ★ Science Park GrazTop 3 finalist · Start-up Idea Competition (Digital Economy & ICT)
- NVIDIA InceptionProgramme member
- Intel Partner AllianceAlliance member
- Google for StartupsCloud Program
- TUM Venture LabsSupported startup
- CyberLab · CyberForumAccelerator participant
- IPAI · KI Festival 2026Exhibitor
- EU AI PactVoluntary pledge signatory
- UN Global Dialogue on AI GovernanceWritten contribution accepted
Two ways in.
Read-only first · Missing evidence stays missing · What works today · More agents and platforms are coming: Notify me
