Your gates see a trusted process on the right account. We see the agent.
AI agents act on every laptop with your employees' access, and every gate you own says yes. The PreAI suite judges what AI writes, what it can reach and what it does, on the device, before it runs.
Twelve gates face outward. The AI agent starts inside them.
Each gate was built to catch a stranger, malware, or data going to a forbidden place. An AI agent is a program you installed, running as your employee, on a compliant device, to destinations you allowed. Every gate lets it through, correctly. At each of those points the suite asks a different question.
| Your gate | What it asks | What the suite asks instead |
|---|---|---|
| DDoS protection | Is a flood arriving from outside? | Is one AI session running away? Quarantined until a person looks. |
| Network firewall and IPS | Is the destination allowed? Is this a known attack? | What is inside the prompt, and should it leave? Credentials redacted, personal data as placeholders. |
| Web application firewall | Is a visitor attacking our website? | Does anything the AI reads contain orders aimed at it? Stripped, and flagged as data. |
| Internal and host firewalls | May this machine reach that server? | Should this program do this action, right now? Judged before it runs. |
| VPN and Zero Trust | Who is asking? | Which program is acting for them, and what may it do? One actor per agent, secrets never shown to it. |
| Email gateway | Is incoming mail dangerous? | Is confidential data leaving by mail, and to whom? Inside the company only. |
| Web proxy | Is the site allowed? | What is being pasted into the AI site? Placeholders instead, found before the send. |
| EDR and antivirus | Is this process behaving like malware? | Is this trusted program about to do harm, and can we undo it? Stopped, or rolled back. |
| Network access control | Is this device allowed on? | Which AI is on this device, and is each one guarded? Inventory, coverage, proof. |
| SIEM and SOC | Do the alerts add up to an attack? | What did the AI do, and can anyone prove the record is intact? Kept as evidence, on a SOC you run. |
The gates ask about the outside. The suite asks about the inside: this trusted program, this action, this data, right now. Your twelve gates stay exactly where they are.
Every one of these passed every gate.
Three public incidents from the last year, in the words of the people who lived them.
The agent said sorry.
An AI coding agent deleted a live production database during a declared code freeze, then said rollback was impossible. It wasn't. An apology is not a rollback.
July 2025It takes orders from strangers.
A rogue commit put a data-wiping prompt into an AI coding extension used by hundreds of thousands of developers. It shipped. Anyone who can write can give it orders.
July 2025You gave it the production keys.
In turbo mode, asked to clear a cache, an AI editor's agent wiped the user's whole drive past the recycle bin, then apologised. Day one, production access, no supervisor.
December 2025For the people who will be asked what happened
You enforce least privilege for people. Your agents run on the developer's.
The attack arrives as a paragraph. There is no signature for a paragraph.
Every README your agent opens is a script it might run.
A confession is not an audit trail. The record is.
Sorry is free. The restore isn't.
A new kind of worker on every laptop. It needs a supervisor.
A no before. An undo after. The record in between.
Each runs on the device. Nothing leaves it to be checked. Use one, two or all three.
PreAI Sentry
Inside the coding tools. Leaked keys, personal data, risky packages and hidden instructions caught as the assistant writes, before the commit.
preai.dev →
Tokenvoy
Agents get access, never your keys. The secret goes into the program, not the agent, and is blanked from anything the agent prints.
tokenvoy.preai.dev →
Prahari
Every action judged before it runs. Files put back if it ran. One emergency stop for every agent on the machine.
prahari.preai.dev →The auditor reads the record, not the apology.
Each product reports to a server you run, on your premises or in your cloud. Evidence goes up, signed policy and commands come down. No product server sits in the AI path.
Team Server
Policy for every developer machine, findings in one place, a gate in the build pipeline. Your code never leaves.
Central vault
Shared vaults for teams. Access for the agents, keys for nobody but the people you choose.
Enterprise SOC
The whole fleet on one console: alerts, triage, commands. Nothing loosens without two people approving.
Policy & Compliance
Your rules become policy. Every decision carries the obligation it supports, ready for the auditor.
Evidence an auditor can verify was never edited. Plugs into the identity, device management, SIEM, code hosting and ticketing you already have.
See it on your own machines first
A short pilot on a team you choose, before anyone commits.
- Learn
Nothing blocked, everything seen
The suite watches what your agents actually do, and shows you what it would have stopped.
- Protect
Risky actions stopped
Switch to blocking. Each stop comes with a plain reason, and only a person can allow it.
- Decide
A report, then your call
A written summary of what it found. Walk away and clean removal puts every file back.
Enterprise
The suite, under one agreement
Custom
All three products and their enterprise servers, quoted to your seats and scope.
Four questions people ask first
We have EDR, DLP and a proxy. Why this?
They watch for malware, for data leaving, and for forbidden sites. A trusted program doing allowed things under the right account trips none of them. The suite sits where the agent acts and judges the action itself, before it runs.
Does anything go to your cloud?
Only identity, licences, updates and signed feeds. Every check runs on the device. Findings, secrets and records go to servers you run, or stay on the device if you run none.
Do we need all three?
No. Each is complete on its own and shares no server with the others. Start with the risk that worries you most. The suite is for one agreement and one security team watching the whole fleet.
Will it catch everything?
No, and we would rather say so. AI tools change every week, detection is pattern-based, and rollback is best effort inside other programs. The suite is a strong floor, not a guarantee. Keep version control, backups and the security you already have. Our disclaimer says this in full.