
HBX Access · Agent Access
Put your AI agents to work on your legacy systems, with no permanent password.
Your identity provider stays. HBX lets the agent into one system, for one job, in the name of the person who assigned it. Reads run at machine speed; anything destructive waits for a named person to say yes.
The problem
Your agent reaches production with a password nobody owns.
No name behind it
The agent runs on a shared login copied into a settings file. When something breaks, the log names the login, not the person who sent the agent.
Reading and destroying look the same
The login that lets the agent read a record also lets it delete the whole table. Nothing stops at the irreversible step.
Nothing ever expires
The access outlives the job, the project and the engineer who set it up. Nobody remembers to take it back.
How it works
A named person approves. The agent connects straight to your system.
The agent asks, and a named person sees the exact statement on their own device before anything irreversible runs. Reads never wait. The traffic goes straight from the agent to your system.
None of this is invented here: it runs on published standards, so the decisions stay in your identity provider. The standards, one by one.
What you get
Access that asks first, and ends when the job ends.
The approver reads the action
Not a ticket title. The exact action, the target, how much it could affect, the access it would get and how long it would last, on one screen, while the action is still pending. A person can only be accountable for a decision they were given enough to make.
Silence is a denial
The request itself is on a clock. If nobody decides before it runs out, it is denied automatically: the agent is refused, not queued. Nothing sits pending overnight waiting for someone to approve it half-read in the morning.
When the window ends
The access stops working and the path it opened closes with it. Work already done is not undone (no access control can promise that), but the next action doesn't get to start, and the same approval can't be used to get back in.
Nothing is left standing on the target
The login exists on the system only for the window: created for that one job, with only the access it needs, and removed when the window ends. No key is stored beside it, and nothing is written into a settings file. There is nothing for the next person to find.
Copying the token gets you nothing
The access is tied to a key only the agent holds, and every request carries a fresh proof signed with it. A string copied out of a log or a prompt arrives without that proof and opens nothing at all.
The record outlives the session
Who delegated, what was requested, who approved and how long they took, what access was approved, which target it reached, when it ended. Each entry is locked to the one before it, and the agent that generated the record has no way to go back and edit it.
Use cases
Where agents reach production today.

Give us thirty minutes and bring your worst delete.
Name the system you'd never let an agent change. We'll walk you through what HBX Access lets it do, what it stops, and who has to say yes first. If people have to reach the same systems, that is the same engine with a console on the front of it: see HBX Human Access.
