Questions
Everything people ask us about Cognilox.
Plain answers, no fine print. If yours is not here, ask us and we will add it.
Data agent wants to
Delete every row in a table
The basics
Cognilox is a self-hosted verification and defense layer for AI agents, made by Aptibit Technologies. Before an agent's action takes effect, such as changing code, data or infrastructure, or replying to a customer, Cognilox checks it against your company's rules, tests risky database and infrastructure changes first, and then allows, fixes, blocks or holds it for human approval. It also stops hijack attempts and leaked secrets as they happen.
A filter looks at text and guesses whether it is harmful. Cognilox sits in front of what the agent actually does: the command, the database change, the tool call, the reply. It checks that action against rules your company owns, tests the risky ones first, and keeps a signed record of every decision. It can also hold an action for two people to approve, which a filter cannot.
No. Cognilox runs on your own servers, including fully offline sites, and none of your data goes to Aptibit. Its records keep the decision, not the content of what agents said. If you prefer not to run it yourself, Cognilox Cloud is also available.
Agents that write and run code, agents that change databases or infrastructure, agents that reply to customers, and agents connected to your tools and internal systems. Most agents need no code change: you point them at Cognilox and it checks what passes through it.
Safe actions go straight through. Cognilox can start in watch-only mode, so you see what it would have stopped before it stops anything. When it does block, it tells the agent why, so the agent can correct itself.
Yes. Your company writes and owns every rule, in a form your team can read. There are no hidden rules, and a change that would remove a protection is refused before it takes effect.
Book a live demo. We will show Cognilox stopping real attacks on real agents, then talk through your agents and how you would like to run it.
Going deeper
When most of an action is fine and one part is not, Cognilox removes the problem and lets the rest through. For example, it takes a secret, a payment card number or an unsafe link out of a reply. When it cannot fix an action safely, it tells the agent exactly what to correct.
It waits. The people you choose are asked to approve it, two of them if you want, and the agent's own owner can be kept out of the decision. The approval covers only that exact action, so a changed or repeated action is checked again.
Hijack attempts hidden in documents, web pages and tool results, secrets leaking out in requests or replies, attempts to smuggle data out or reveal an agent's private instructions, and tools that change behind the agent's back or do not exist. Decoy credentials expose an agent that has been taken over, and Cognilox isolates it.
Payment card numbers are removed by default. Other personal data is stopped under rules your company sets, so you decide what may leave and where.
Yes. One switch stops one agent, one team or every agent, and another lets them carry on. You can also cap each agent's calls, usage and spend.
On your own servers, fully self-hosted, or in a fully air-gapped site with no internet connection at all. Cognilox Cloud is the third option, for teams that would rather not run it themselves.
There are Personal, Team and Enterprise plans, plus self-hosted, air-gapped and partner editions. Pricing depends on how many agents you protect and where you run Cognilox, so talk to us and we will quote for your setup.
Aptibit Technologies, an AI company founded in 2024 in Kolkata, India. The same team builds Visylix. We built Cognilox because we needed it: it already checks the actions of our own AI agents before they take effect.
See it stop an attack, live.
Thirty minutes, real agents, real attacks. We show you what Cognilox allows, fixes, holds and blocks, then talk through your own agents.