We build the AI offensive security harness for red teams
Our agent works inside the scope you sign. It takes one step at a time, stops before anything risky and writes every move into a record you keep.
01/ 06harness
The harness, in four parts
A model can plan an attack by itself. We build what sits around it: where it may go, how it moves, when it stops and what it writes down.
- 01read the scope file0:02
- 02map the login flow1:14
- 03list session cookies0:41
- 04replay an expired token0:09
- 05edit a test recordheld
Step 05 would change one record on account-api.
The run is paused. Nothing is sent until you choose.
aapprove oncesskipvview the request
02/ 06runs
Every run, logged as it happens
Follow a run line by line: each step, each request we refused and the moment it stops to ask you.
Why a harness
An agent that attacks systems on your behalf should never run loose. So we put it in a harness: a written scope, one step at a time, you before anything risky, and a record of all of it.
03/ 06approvals
You approve the step that changes data
When a step would write, delete or restart something, the run stops. You read what it will send, then approve it once, skip it or tighten the scope.
Edit a test record
- host
- account-api
- held for
- it changes data
- sends
- 1 request
- then
- step 06 restores the record
04/ 06record
A record you can file
Every run ends as a plain log: what the agent did, what we refused and who approved what. You keep it, and it reads the same in your report.
runs/0412.log, as printed
- 12:04:11youstarted the run
- 12:04:13agent01 read the scope file
- 12:04:15agent02 mapped the login flow
- 12:05:40scoperefused admin-portal, out of scope
- 12:05:41agent03 listed session cookies
- 12:06:11agent04 replayed an expired token
- 12:06:20gate05 held, changes data
- 12:11:47youapproved 05, once
- 12:11:49agent05 edited one test record
- 12:11:52agent06 restored the record
- 12:14:31youclosed the run
05/ 06teams
We build Heedio for people who test with permission
- a
Red teams
Long engagements, where every agent step has to land in your report.
- b
Penetration testers
Client work with a written scope. Our agent touches nothing it leaves out.
- c
Product security
Your own staging systems, where you approve any step that changes data.
- d
Security researchers
Targets you own or may test, one agent step at a time.
We don’t build for anyone without written permission to test the target.
06/ 06questions
Questions
Ask us anything we haven’t answered here.
What is an AI offensive security harness?
It’s the software we wrap around an AI agent doing offensive security work. We hold it to your scope, run it one step at a time, stop for your approval and keep the record.
How is that different from a scanner?
A scanner runs a fixed list of checks. Our agent picks its next step itself, so we keep every choice inside the limits you set.
Can I use Heedio today?
We’re opening access in stages. Request access, tell us what you test, and you’ll hear from us first.
Which models and tools do you use?
We’ll walk you through our stack when we open your access.
What will it cost?
We’ll share pricing when we open access.
What may I point it at?
Only systems you own or have written permission to test, and only what your scope names. We hold you to that.