Test us before you meet us
Free to download and inspect — 4.8k stars, public source, six published papers. No demo request, no form, no call. Hand it to your engineer and let them try to break it this afternoon.
AGNT Forward Deployment
Clear the backlog.
Keep the team.
We put AI agents alongside your specialists to do the collecting, checking and chasing. The same team moves more work, faster. What we can promise is that the work we take is paperwork, never judgment — the rest is a decision only you can make, and we will help you make it in the open.
The software is public. Have your IT lead download it and try to break it before you call us.
Individuals at these organisations use AGNT
Stated plainly: these are individual practitioners who use the software, not company-wide deployments, partnerships or endorsements. Ask us on the call and we will tell you exactly which is which.
What Changes
The work we take is the paperwork.
Most expert work isn't expertise. It's collecting, checking, formatting and chasing. Agents take that part. Your specialists keep the judgment — and stop spending their day earning the right to make it. What happens to headcount afterwards is your decision, not ours, and we would rather say that on the first page than let you discover it later.
The task
How it works now
With agents doing the legwork
Building a case file
Your specialist spends most of the day chasing documents and re-keying them
The file is assembled and flagged before they open it. They decide.
Checking documents
Spot-checked, because reading every page isn't physically possible
Every page read. Your people are handed the exceptions.
Watching for problems
Reviewed weekly by hand, and out of date by the time it's read
Checked continuously. Someone is told the moment it matters.
We are not going to print a multiple next to those rows. Your real numbers depend on your process, and we measure them in week one before anyone claims anything.
This page answers questions from four different people. Skip to yours.
Who you would actually be working with.
You are not hiring a firm and you will not be handed to a delivery team. One person maps your process, builds the thing, and answers the phone when it breaks.
One person. Which is the point, not an apology.
You can read everything they have built before you meet them. The software is public and you can go through it line by line. The research is published with the working needed to reproduce it, including the results that went the wrong way. Very few people selling you AI work will let you audit their thinking before the first call.
The trade is honest: you get the person who built it rather than the person who sold it, and in exchange we can only run two or three of these at a time. That is why the first call is a real conversation about whether your work is a fit, not a pitch.
The software isn't new. The service is.
We are early as an operating partner and you should price that in. What is not early is the thing we run your work on — it has been in production for years, in the open, where anyone can check it.
40,000+agent runs against live production tools — not demo environments. We published the field notes, including what broke.
119tools it already connects to, plus 6,000+ more through Zapier. Your stack is almost certainly on the list.
Yearsin production and in the open, before we ever offered to run it for anyone.
Check the privacy claim yourself
Put it behind a locked-down connection and watch exactly what it reaches for. We will hand you the full list of connections it makes before you install anything.
No account executive
The first call is with the person who would do the work. If we can't answer your questions in that call, you have learned something for free.
Where does the freed-up time actually go?
The fair question, and the one most vendors skip. If your volume is set by the market rather than by how fast your people work, then capacity you free up has to go somewhere. Here is where we have seen it go — decide which of these you want before you start.
The queue that never clears
Most teams are behind before the week starts. The first thing extra capacity does is clear the work already waiting — which shows up as cycle time, not as spare people.
The checks you skip today
Every team samples because reading everything is impossible. Capacity spent on complete checking instead of spot checking is quality your regulator and your customers can see.
Volume without a hiring round
If you're growing, this is the year you don't add six people to keep up. That is the saving — the hire you didn't have to make, not the person you let go.
If none of those three is true for you, be honest with yourself about what you plan to do with the capacity — your team will work it out on day one, and they will be right.
Nothing goes live without a signed people plan
Which roles change and how. Where the people currently doing that work end up, by name and title. Who owns their retraining, from both sides. What the productivity baseline is for the next two review cycles, so nobody gets quietly re-targeted six months later. And what happens to base pay, banding, bonus and overtime for every person named — including if the role is re-graded later — signed by the executive who owns that budget and countersigned by the manager whose team it is. Both signatures, or it is not signed.
We will not write your compensation policy — that is yours, and a promise from us would be worth nothing to your staff. We will refuse to go live until someone with the authority to answer it has, in writing. It is a gate, not a document. If it is not signed, the switch does not flip — including on the second process, and the fifth, not just the pilot. And no name appears in it before that person has been told in person by their own manager.
The plan is yours and it is written with the person who runs the team, not about them. We will not promise your staff anything on your behalf — it is not our payroll, and a promise from us would be worth nothing to them.
When it gets something wrong.
It will. Any system that touches enough work eventually mishandles some of it, and a vendor who won't discuss that in advance is one you find out about later. Here is how the exposure is contained.
A person signs
Agents assemble, check and flag. The determination stays with your qualified staff — the same person who signs it today signs it after.
Doubt escalates
Below a certainty level it stops and hands the work to a person rather than guessing. We set a deliberately cautious starting point, show you how much it routes to humans, and tune it with you in week two — you are not handed a blank dial.
Every file has a receipt
What was read, what was skipped, what was flagged, who approved it. Exportable and readable without our software, so it is still evidence in four years when we are not involved.
Hard stops you set
Actions that cannot be undone require a signature. Spending and volume caps are enforced by the software before an action runs — not written into an instruction the model could talk itself around.
Those are the controls. The other half is the contract — indemnity, liability limits and what our insurance covers — and that belongs in an agreement your counsel has read, not on a web page. Bring them to the first call. We would rather have that conversation early than discover the gap at legal review.
How we work.
Four stages. Each has a fixed scope and a date, agreed before it starts. Stop after any of them and you keep something that works.
Operations Mapping $35,000 fixed
You end the three weeks with a working agent running against one of your real processes, a measured before-and-after on that process, and a written go or no-go. Not a slide deck. If you continue, the fee comes off the next stage. If we say no-go, you keep the agent and the measurements anyway.
2–3 weeks · fixed fee, agreed before we start
What it costs you, besides the fee.
Roughly 6–8 hours of one process owner across three weeks, in short sessions rather than a workshop. About 2 hours from whoever administers the systems we connect to. Nothing from the wider team unless you want them involved. If your best person is unavailable for those hours, tell us on the call and we will schedule around it — mapping a process without the person who knows why the exceptions exist is how these engagements produce confident nonsense.
Build & Deploy
Set up on your systems and connected to the tools your team already uses. Approvals, limits and undo set to your rules. Training and written guides included.
Fixed scope, fixed dateWe Run It
We watch it and take the call when something breaks, so it never becomes your team's problem. New agents as the work grows. A monthly report your board can read.
RollingYou Run It
Your people take it over. We provide updates, support and someone to call. Works completely offline if that's what you need.
AnnualStage one is a fixed fee so you know exactly what the decision costs. Everything after it is scoped and quoted once we understand the work — one number, in writing, before you commit to anything. We do not do change orders you did not see coming.
You're not commissioning custom software.
The question your technical people will ask. We set up a product that already exists — and keeps existing whether or not we work together.
The product
AGNT. A real product with its own version history, documentation and public source. You can download it today without talking to us.
Your setup
That product, arranged around your work. The same building blocks everyone else uses — not a one-off written only for you.
If a piece is missing
We add it to the product. No private version you end up owning and paying to maintain.
Nothing becomes your problem
No custom system that one contractor understands and your budget maintains forever.
Each project speeds up the next
What we build for one client becomes part of the product. Your second phase costs less than your first.
We run our own company on it
Problems hit our work before they ever reach yours.
Everything it connects to.
Published here rather than sent on request, because your security team will ask and the answer should not depend on a sales call. Every one of these is optional except the tools you deliberately connect.
| Reaches out to | What it sends | When | Turn it off? | If you block it |
|---|---|---|---|---|
| Your AI provider Only if you choose a hosted model |
The content being worked on | Each time an agent runs | Yes — run a local model and this disappears entirely | Agent runs stop cleanly and report why. Nothing queues on disk. |
| The tools you connect Your mail, files, database, ticketing |
Whatever that task requires | Only the tools you approve, only when used | Yes — you grant each one individually | That one tool is unavailable. Everything else carries on. |
| Software updates Our server |
Your version number and the IP it came from. No customer data. | On start-up | Yes — offline installs are supported | Nothing. It carries on at the version you have. |
| Add-on catalogue Our server |
The package name you asked for, and the IP it came from. No customer data. | Only when you browse or install one | Yes — never contacted if you do not use it | Nothing. You simply cannot browse add-ons. |
| Us — your work | None of it. Ever. | Never | Nothing to turn off — no path exists for it | Not applicable. To be exact: the two rows above are our servers and they see an IP and a version string. Switch both off and we learn nothing about you at all. |
Your security team gets the exact hostnames, ports and failure behaviour in writing before the first meeting, not before install — and they are welcome to verify it themselves, since the source is public. If they find something on the wire that is not on this list, we want to know before you do.
Your information stays under your control.
The questions that kill most AI projects are answered before the first meeting.
🏢
It runs on your computers
Your systems, in your building or your own cloud account. We hold no copy of your data, ever.
🔑
Your accounts, your control
You pay AI providers directly and we never mark it up. It is the one cost that moves with volume, so we model it against your real numbers in stage one — a floor and a ceiling, before you commit. Run a local model instead and it goes to zero.
📋
Everything is recorded
Every action is logged — what it did, what it cost. Audits take minutes, not weeks.
👤
We hold no keys
Our access uses accounts you create and revoke, logged in your systems. No standing credentials, no permanent remote connection, and none at all after handover.
🚪
You can end it any time
Switch off our access and keep running. It doesn't stop working when we do.
What happens when it ends.
Every vendor tells you what it costs to start. You should ask what it costs to leave — and be suspicious of anyone who has not written it down before you ask.
30 days, both ways
No minimum term on the running stage, no automatic renewal, no penalty for stopping. If we are worth keeping you will keep us.
All of it, in a usable form
The configuration, the prompts, the connections, the measurements and the records — exported in open formats another firm can pick up. Not a login to something of ours.
Nothing
Handover and transition support are included in the final month. We do not charge you to leave and we do not hold the configuration hostage.
You keep running
The software is public and self-hosted — it does not stop when we do. Your team can operate it, or another firm can. That is the point of not writing you a private system.
We are early, so you are right to ask the last one. It is exactly why nothing we build for you depends on us being here.
We are not right for everyone.
Small team, two or three projects at a time. That is a real limit, not a sales tactic — and it means we would rather tell you no on the first call than take your money and discover it in month six.
Worth a call
Let's talk
- Sensitive information, and off-the-shelf AI is already off the table
- One process you can name, with real volume behind it
- Someone who owns the outcome and can sign
- Happy to start in one department and expand
Not worth either of our time
Probably not us
- “We need an AI strategy” with no process named
- Looking for the cheapest bid
- Endless custom development under another name
- Decisions that shouldn't be handed to software
First-call questions.
Usually in this order.
What does it cost?
Stage one is $35,000. Fixed, not a starting bid, so you know what the decision costs before you make it. What comes after depends on the process and the systems involved. Fair challenge: if we can credit stage one against stage two, we clearly have a range in mind — so here it is. A single process, end to end and running, has landed between roughly six figures and the low hundreds of thousands, driven mostly by how many systems it has to touch and how much sign-off sits in the middle. We will not pretend that is a quote. We will give you one number in writing after stage one, and we do not raise it with change orders you did not see coming.
Will this cost people their jobs?
That hasn't been how our clients use it. Agents take the gathering and checking; your specialists keep the decisions and handle more volume. If your goal is cutting headcount, we're probably the wrong partner.
Who else have you done this for?
Straight answer: we are early as an operating partner, and you should price that in. The software is a different story — people at Howard Hughes, IntelyCare, CPower, Comarch, Bishop Fox and Zapier use it, and we have published field notes from 40,000 agent operations run against real production tools. So we are asking you to judge us on a product you can download and inspect today, plus a first stage small enough to be a cheap test — not on a reference list we do not have yet.
You are a small team. What happens if you disappear?
Fair, and the reason it matters less here than with most vendors: what we build is a configuration of a documented product, not a private codebase only we understand. Your team is trained on it, the runbooks are yours, and stage four exists precisely so you can take it over. If we vanished tomorrow you would still be running.
What do I tell my board?
That you have a named process, real numbers, a fixed scope and a date. After stage one you can show them working software — not a request for budget to explore AI.
Where does our information go?
Never to us. The software runs on your systems and we hold no copy of your data. If you use an outside AI provider, that traffic goes from you to them under your own contract — we configure the limits, you own the relationship. If that is still too much, it runs fully offline against a local model and nothing leaves the building at all. We will show you the exact list of connections it makes before you install it.
Do you guarantee results?
We agree what success looks like up front, and we fix anything that misses it. We don't guarantee revenue or perfect accuracy. Anyone who does is describing a sales pitch, not engineering.
How quickly can we start?
Stage one runs two to three weeks and begins when capacity opens. Because we hold only two or three projects at once, start dates are genuinely limited.
Bring us one process.
Keep your people.
Thirty minutes is enough to tell whether this is worth either of our time. Bring the person who runs the work — they will spot the problem faster than we will.