The world’s most complete and intuitive AI Governance Infrastructure.
Built for enterprises, governments, developers, AI builders and personal AI users. Govern, verify and trust every AI system from one unified infrastructure. Start with one capability. Grow into the complete platform without replacing anything.
proof/provenance.json
- We published
- f042bf6c191f12bd…41c16ff061febb25
- Your browser computed
- not checked not checked
Recomputed in your browser from the bytes you received. Nothing was sent anywhere.
- Actions
- Start now Contact Sales
- Runtime
- Sentinel Runtime
- Developers
- Developers & AI builders
- Solutions
- Enterprise Government
Your personal AI agents, on the same platformLearn more
One runtime. Every interface. All verifiable.
Adopt the layers in any order. One operation is computed once and read by every surface above it, so nothing you adopt later replaces what you adopted first.
- The operation
- What an AI system actually did, as it happened.
- Evidence Engine
- The envelope the operation became, and the receipt issued for it.
- Trust Intelligence
- Compliance, risk and quality composed into one explainable posture.
- Surfaces
- Executive, Evidence, Passport, Certification, SLA and Benchmark, all reading the same record.
- Export
- Portable, verifiable offline, with no lock-in.
One platform, seven products, one record between them. Evidence Engine, Flight Recorder, Memory Receipts, Agent Passport and Trust Intelligence produce and read that record; Sentinel Runtime governs the execution that creates it; VerifAIer Me is the same infrastructure at the scale of one person. They are not seven tools that integrate. They are one platform you adopt in parts, and each destination states its own maturity when you open it.
Every claim on this page ships with the file that proves it.
proof/provenance.json
- We published
- f042bf6c191f12bd…41c16ff061febb25
- Your browser computed
- not checked not checked
Recomputed in your browser from the bytes you received. Nothing was sent anywhere.
The digest beside it was recomputed in your browser from the bytes you received. Change one character of the file and it stops matching.
- What this proves, and what it does not
- Read the boundary before you rely on it.
- How the digest is produced
- SHA-256 over the served bytes, in the page.
ev_bf6442d03d1d461589a3b708bbd19d28
- Provenance
proof/provenance.json· Where the operation came from.- Evidence Engine
proof/evidence-envelope.json· The envelope the operation became.- Evidence receipt
proof/receipt.json· The receipt issued for that envelope.- Trust Intelligence
proof/trust.json· The posture derived from it.
Show the remaining 5 files
agent passport · compliance · quality · risk · signals
- Agent Passport
proof/api-passport.json· The trust document that cites it.- Compliance
proof/compliance.json· Control coverage read from it.- Quality
proof/quality.json· Quality signal derived from it.- Risk
proof/risk.json· Risk signal derived from it.- Signals
proof/signal-export.json· The export that carries it outward.
One evidence layer means the surfaces above do not keep separate records. They read this one. Sentinel verdicts and Flight Recorder sessions are captured separately.
9 files · one identifier · each verifiable on its own
Deploy VerifAIer in minutes.
No account. Nothing to configure. The first assessment finishes before you have finished reading this sentence.
- Start
One click. No account, nothing configured.
- Assess
Choose a scope. The run finishes in well under a second.
- Read the result
What it examined, what it did not, and the two questions it refuses.
- Open the evidence
Every answer traced back to the file that produced it.
- Trust it
Verify a receipt yourself, offline, with the network off.
- Operate
Governance becomes the thing that runs, not the thing you file.
The full path · start → assess → read the result → open the evidence → trust it → operate
Start the guided pathSee the full pathThe agent is watched, not trusted.
A flagship product in its own right. It runs beside an AI agent, on your machine, and keeps the record you will be asked for.
- Sentinel Runtime
- Sees the operation before it happens, and can refuse it.
- Memory Receipts
- Sealed receipts prove content is unchanged since it was kept.
- Flight Recorder
- A deterministic, ordered event list, readable in order long after the run.
- Continuous
- Always on, quietly, with nothing leaving the machine.
Sentinel sits beside an AI agent, records what actually ran, and refuses to guess when it cannot tell.
- Sees the operation before it happens
- Records what actually ran
- Refuses to guess when it cannot tell
- Keeps everything on your machine
VerifAIer does not claim to prevent every unsafe change, guarantee secure code, or support every IDE. Evidence arrives through local adapters that read what already happened, not a live provider wire.
Official derivative required. Sentinel is where the mark carries the most weight, so the slot is left empty rather than approximated.
- Claude Code
- Cursor
- Packaged installer
Sentinel installs into the surfaces above. There is no packaged installer, and the download page says so.
Two products- Sentinel Runtime
- Beside the agent, in the editor and the CLI.
- VerifAIer Me
- Your own workspace. Import a conversation you already had to begin. Nothing is captured automatically.
src/billing/invoice.py 41 def issue(self, customer): 42 - retry(times=3) 43 + retry(times=0) 44 return self.send() > agent: applied the change you asked for
Recorded, not judged.
The operation was seen before it ran and written down after it did. No severity is attached, because nothing here ranks findings.
- Seen before it ran
- Left this machine
- Receipt
- Input
- The agent’s request.
- Tool call
- The operation invoked.
- File action
- What changed, as a diff.
- Decision
- The verdict for the step.
- Evidence receipt
- Content-addressed.
Governance where the code is actually written.
2 / 7integrations available today
- Editors
- Claude Code, Cursor
- Terminals
- CLI, coding agents
- Any MCP client
- connects the same way
VerifAIer runtime
Local execution. One operation, computed once.
- Receipt on disk
- open evidence format
- Policy decision
- recorded, not inferred
- Session timeline
- readable in order
It installs into the tool you already have open, runs locally, and leaves a receipt on disk you can read without us.
$ python scripts/install_claude_mcp.py MCP server registered # then, in the editor > audit the change you just made receipt written · readable without us rcpt_711e51de6f00491fb3c82256fdeca7c3
MCP server
python scripts/install_claude_mcp.py
Governed inside the editor. Every audit lands as a receipt on your disk.
MCP server
python scripts/install_cursor_mcp.py
The same server, the same evidence model, a different editor.
Gemini CLI
MCP server
Codex CLI
MCP server
Windsurf
MCP server
OpenCode
MCP server
Continue
pending verification
- Developer SDK
- One typed client over the running runtime. Nine services, no duplicated engines.
- CLI
- The same operations from a command line, on the same local runtime.
- Integrations
- Providers registered, executed and receipted through the runtime rather than beside it.
- Marketplace packs
- Policy packs distributed as metadata. Nothing executes and nothing downloads.
Install
One command adds the MCP server to your editor.
First audit
Ask for an audit of the change you just made.
First evidence
A receipt lands on disk, readable without us.
Regulator-ready evidence, without sending a byte to us.
One runtime, deployed inside your estate. Roles, tenancy and audit trails on top of the same evidence every other surface reads.
- CIO
- One platform to install, one runtime to operate, no hosted dependency.
- CISO
- Air-gapped and sovereign deployment. Verification works with the network off.
- Governance
- Coverage, findings and drift read from evidence rather than asserted.
- AI operations
- Every operation recorded once and readable by every surface above it.
Nothing to flag.
Ready
Twenty capabilities checked. Fourteen answered. Six could not be reached, and are reported as unknown rather than as failures.
14 / 20 answering
A ratio, not a grade. No composite score is computed.
- What needs action
- 6 unreachable
Partly done - What improved
- No comparison yet
Cannot tell - Estate health
- 14 / 20
Ready
- Assess
- Evidence
- Posture
- Report
Sovereign by construction, not by promise.
VerifAIer runs on infrastructure you control. The evidence stays where it was produced, and it can be verified with the network off.
- 3
- policy packs ship today
- 0
- bytes leave your perimeter
- Offline
- verification, network down
There is no government-specific product surface. These are the same packs an enterprise adopts, and filing them under a public-sector heading would tell you something untrue about the product.
Everyone is about to have an AI that acts for them.
Assistants are moving from answering questions to booking, buying, writing and deciding on your behalf. The record of what yours did belongs to you, not to whoever built it.
VerifAIer Me is the personal governance layer for that. Same evidence system as the enterprise platform, pointed at your assistant instead of an organisation’s agents.
Available today: AVAILABLE
Three moves. You start it, and nothing happens until you do.
- Import
A conversation you already had with an assistant. Nothing is captured automatically, ever.
- Seal
Advice, a decision, a result. Sealing binds the time to the content under a deterministic digest.
- Check, later
Any edit to the content breaks the seal. That is the whole promise, and its whole limit.
Evidence that the content is unaltered since sealing, under a SHA-256 proof_id. It is not a judgement on whether it was right, it does not observe what a model retained or reused, and VerifAIer does not claim to solve AI memory correctness.
- title
- Q3 compliance review summary
- type
- advice (level 2)
- content hash
- 6b8bce923fe8…
- proof_id
- dc5b3bd05b46…
- sealed at
- 2026-07-29T22:45:03Z
Designed for the personal-agent era.
Soon people will keep assistants and agents that persist for years. VerifAIer Me exists so that protection can belong to the individual rather than to whoever hosts the model. This is where it is going, and none of it is here yet.
- Personal AI protection
- A trust layer that sits with the person, not the platform.
- What your agent did
- Visibility into the actions a personal agent took on your behalf.
- Evidence you own
- Personal receipts and memory that stay yours if you leave.
- Desktop
- A local application for macOS, Windows and Linux. Planned after VerifAIer v1.
- Apple App Store
- An iOS and iPadOS release. Planned after the desktop application.
- Google Play
- An Android release. Planned after the desktop application.
Not claimed today
No passive monitoring. No automatic capture. No continuous agent protection. No packaged installer of any kind. Nothing here can be downloaded, and no release date exists.
Enterprise and operational. Sits beside AI agents, coding workflows, teams and governed systems.
Individual protection and evidence for your own AI use, evolving toward the personal-agent era.
Same trust system · two products · neither is a smaller version of the other
Audit an AI conversationSee the product direction