A prospect just invited you to their online security portal.

225 questions that need completing, and until they are, nothing on the commercial side moves.

TrustFill puts your existing AI to work inside their portal: the one they sent you, with their login and their quirks. It drafts answers; you accept or correct them. And every correction is written to a journal your agent can use next time, so the next questionnaire starts smarter than this one did.

  • Most of a day becomes minutes Answered in the portal itself. No export, no import, no copy‑paste.
  • It compounds Every correction your team makes becomes precedent. The second questionnaire is already faster than the first.
  • We keep nothing The session stores no data at all. Your journal lives in your bucket, read only while you are working.
  • Minutes to start Sign in, open the questionnaire, press go. There is no integration project.
Book a demo See how it works

Bring your own agent, or the one you are still choosing. You log in yourself, and nothing about the session is recorded.

The problem

Procurement stalls at the questionnaire.

And it stalls inside someone else's web app. Your prospect sends a link, not a spreadsheet, and somebody on your side logs in and works through hundreds of questions by hand. Answer-library tools do not help, because the questionnaire never leaves the portal.

The timeline slips at the last mile

The evaluation is done and the contract is drafted, but signing waits while somebody on your side works through hundreds of questions by hand.

Every prospect sends a different portal

There is no market leader in questionnaire portals. Each procurement suite ships its own, with its own layout, so there is nothing to standardise against and nothing that exports cleanly.

By hand it takes most of a day

Over ninety seconds a question, and that is the good case: you already know the answer and only have to find the field, paste it and check it against the policy. When you do not, it is a search through documents and a message to whoever owns that control. At 225 questions, ninety seconds each is five and a half hours before anyone has queried a single answer.

The flow

How one questionnaire runs, start to submitted.

Attach your agent

An endpoint, a key and a three-line answer contract. Once, not per portal.

Log into the questionnaire portal

You log in yourself, in a live session you can see and drive. The credentials stay with you and the agent never receives one.

The agent answers the questions

It walks the sections, drafts answers with the reasoning attached, and enters only the ones it can source into fields that are still empty.

The officer confirms the answers

Accept what is right. Everything the agent would not enter unattended is waiting in one list, with the reason it is waiting.

Corrections to answers

You correct an answer, revise one a colleague left, or attach the document only a person can attach. Correcting in the portal itself counts too.

The Journal records it as you go

What the agent said, what you changed it to, why, and who settled it. Append-only, and exportable as one Markdown file.

Submit the questionnaire

Always a person's act. TrustFill has no route to the submit button and never will.

TrustFill · session panel
Action required2 218 / 225 answered
114drafted · sourced
Do you have a procedure to change default passwords shipped with software?
Answer: Yes
Standard hardening step in the platform baseline.
AcceptCorrectShow me
224needs a document
Please attach the report of the application penetration test.
Only you can attach this.
Show meAttached, mark done

What the officer sees at step four: 218 of 225 done, two things left that need a person.

Your agent, our surface

We are not selling you another AI. We make yours useful, and smarter each time. It can take the credit.

Maybe you already run an agent. Maybe you are halfway through building one, or still deciding. Either way the gap is the same: an agent with no hands inside your customer's portal, and no memory of what your team decided last time. That is what we are. Connecting one takes about as long as making a coffee — an endpoint, a key, and a three-line answer contract.

And the agent gets better here, not just busier. Every question it is asked arrives with the deal context, your house rules, and the answers your own team already approved for questions like it — ranked, so the same question coming round again outranks a vaguely similar one. An agent that knows nothing about your posture on Monday is answering with your team's own words by Friday.

ANSWER

One of the options actually on the page, verbatim. A value the control does not offer is refused, not approximated.

WHY

One sentence naming what the answer was based on. Stored with the answer and shown to whoever reviews it.

BASIS

sourced, inferred or unknown. Only sourced is ever entered unattended.

Only the agent you designate may draft an answer. Nothing else can — not our portal, not the assistant you type to, not a script with your API key. That boundary is enforced, not promised.

Your whole compliance team can work in it. Everyone reviews, corrects and signs off in the same place, and every review, edit and correction is recorded against the person who made it. That record is visible only to your administrators — it is your team's working history, not something we look at and not something your customer sees.

Your agent, your corpus

We don't build the intelligence. That's deliberate.

Answering a questionnaire well means reading your policies, your contracts, your knowledge base, and what your team said last time. Handing all of that to somebody else's model is a compliance decision most companies won't want to make, so we don't ask for it. You connect your agent; we put it inside the portal, and hand back what it learned. We tried building the agent by hand first, and the cleaning took longer than the agent did, which is why the export is structured from the start.

The Journal

Every correction comes back to you as a file.

When you change an answer, we capture the question, what the agent said, what you changed it to, and why — then hand the whole questionnaire back as one Markdown file. Feed it to your agent and the next questionnaire starts from what you taught it on this one.

northwind-bank-vendor-security-assessment-journal.md
# Northwind Bank — Vendor Security Assessment

- **Portal** https://portal.example-grc.test/assessments/8841
- **Attempts** 3, first opened 2026-09-08 09:14
- **Questions recorded** 225
- **Answers a person changed** 12
- **Answers carrying a reason** 198
- **Documents requested** 8
- **Exported** 2026-09-10T08:22:15Z by TrustFill

## What this file is for

Every answer given on this questionnaire, with who gave it and why. Feed it to
the agent that answers your security questionnaires: the entries marked
**corrected** are the ones that teach it something, because a person looked at
the agent's answer and disagreed. Nothing here is inferred - each line is what
was actually recorded at the time.

## Every question

**52** Is code review performed for the application?
- Options: Yes · No · Not Applicable
- Drafted by your agent: **Yes**
- Final answer: **Yes**
- Why: Pull Request Policy mandates review on every PR, two human reviewers, and
  all tests passing before merge.
- Status: accepted as drafted

**107** Are access rights revoked within 24 hours of termination?
- Options: Yes · No · Not Applicable
- Drafted by your agent: **Yes**
- Final answer: **No**
- Why: We do revoke, just not within 24 hours. The SLA is two working days.
- Status: corrected by a person

## Documents this questionnaire asked for

- **16** Infrastructure diagram, and a data flow diagram
- **68** Disaster recovery plan

To be plain about it: we do not retrain your agent for you. Every agent is built differently and reaching into yours is not our business. We produce the intelligence, in a form a person can read and an engineer can wire in an afternoon.

Any portal

The first time anyone opens a portal here, TrustFill learns it. For everybody.

A questionnaire portal is somebody else's web app, and no agent is good at those by default. So when a portal is opened inside TrustFill for the first time, we read it: how it lists its sections, how it words its own progress counter, which marker means a row is open, where an explanation box sits relative to the choice it explains. Then we use the Webfuse APIs to build that portal a set of custom tools, fitted to what we just found.

Your agent never sees any of it. It asks for the questions and gets back numbers, text, options and current values; it sends one answer with an approval. That contract does not move, whichever portal is underneath. What changes is how cheaply and how quickly the portal answers, and that is our problem, not yours.

You open the questionnaire Your prospect's portal, your login, in the session. Even one they built themselves.
We fit it with tools Seconds to about a minute, in the background, while you get your coffee. Nothing is written to the page.
Only ever once That portal is known from then on, and known to every account that meets it next.

That last one is the part worth dwelling on. These tools are not built per customer, they are built per portal. Somebody else's first questionnaire on OneTrust is the reason yours starts fast, and yours on a portal nobody has seen before is the reason the next company's does. The library only grows.

What the tools are actually for

Speed, and tokens. A generic agent has to read the page to find out what is on it, and an enterprise portal's page is enormous. The custom tools hand back the questions as data instead. Same questionnaire, same 32 sections:

Questions on the questionnaire, across 32 sections 225measured
Questions actually present in the page at any moment 17measured
Reading one section as data: text, options, values, types ≈ 1kderived
A page snapshot of one section of a portal like this ≈ 15–40kestimated
≈ 33k tokens to read the whole questionnaire through custom tools
≈ 0.5–1.3M reading it by page snapshot instead, which is 15 to 40 times more, on the reading alone

Two honest notes, because a number without them is marketing. Reading is not where most of the tokens go — every question still goes to your agent with its options, your context and the precedent behind it, and that cost is the same whichever way the page was read. What the tools do is turn the reading from the dominant cost into a rounding error. And the stronger claim needs no arithmetic at all: without tools built for the portal's shape, 208 of those 225 questions cannot be reached at all. Not expensively. At all. Only the open section exists in the page.

Portals and formats

Answered in production, or ready for a first open. The ones marked learns on first open have not been through TrustFill yet, which is a statement about our history rather than about the portal.

RiskProfiler · in production SIG / SIG Lite · answered in production CAIQ · answered in production OneTrust Whistic ProcessUnity Archer Prevalent Venminder Panorays SecurityScorecard UpGuard CyberGRX Aravo ServiceNow VRM SafeBase Conveyor Vanta Drata CAIQ Lite HECVAT VSA Core / VSA Full Something your customer built themselves

Send us a link on the first call and we will open it together. If a portal turns out to be a bad fit we would rather say so than find out alongside you.

Book a demo

Bring us the questionnaire you are dreading.

We will work it with you on the call, in your prospect's portal. Thirty minutes, and you keep whatever gets answered.

We will only use this to arrange the call. The last question decides whether we are useful to you yet, so please answer it honestly.

Questions

You are a security team. Ask us the hard ones.

How is this different from Responsive, Loopio or Whistic?

Those are answer libraries, and they are good ones — but the work stays manual. You still open the library, search it, judge which stored answer fits, copy it, switch to the customer's portal, find the question, paste, and repeat. Three hundred times. We go the other way: into the portal your customer sent, with your agent reading the questions and drafting in place, and you accepting or correcting rather than searching and pasting.

Does the agent get our credentials?

No. You log into the portal yourself, in the session. The agent acts inside that session and never handles a password or a token of yours.

Can it submit a questionnaire on its own?

No. It fills answers it can source into empty fields and queues everything else. Submission is a human action, every time.

Where does our data live?

In your own S3-compatible bucket, which is what we recommend. You enter your bucket credentials, your journal is written there as you work, and once a questionnaire is closed our copy is deleted. We hold the questionnaire you are actively working on and nothing else, because a live session has to be transactional. You can choose to let us keep it instead; we will ask, and we will tell you plainly what that means.

What about the session itself?

The live browser stores nothing at all. No screen recording, no page archive, no keystroke log. Work in it for six hours and there is no six hours of anything to hold — and your customer's questionnaire is never copied out of their portal.

Can we pick a questionnaire up weeks later?

Yes. The answers are where your prospect's portal left them, and the history (the Journal) comes back out of your bucket, loaded for as long as you are working and then let go. Tomorrow, next quarter, next year: nothing of yours has to sit with us in between.

Who can see our team's edits?

Your administrators, and nobody else. Every review and correction is recorded against whoever made it, agent or person, and that is your working history. It is also why we recommend pointing the Journal at your own bucket: the record then sits under your access control and your retention rules rather than ours, and moving it later is a credential change rather than a migration. Your prospect never sees it, and it is not something we look at.

Can it attach or upload evidence too?

If your agent has been built to do that, yes. What the Journal keeps is the reference rather than the file: which document was attached, to which question, by whom and when. That is what an auditor needs, and it leaves your evidence where it already lives, under your own controls, instead of copied into a second system. Where only a person can attach something, the question goes onto your list rather than being guessed at.

Who is accountable for a wrong answer?

Your reviewer, and the record says so. Every answer names who gave it — your agent or a person — with the reasoning attached and nothing ever rewritten.

Are you certifying anything?

No. We are not an auditor and we do not assert compliance. We make answering faster and the record better.

What if we do not have an AI agent yet?

Then you are exactly who this is for. There are capable agents available today — in-house, from a vendor, or one of the general ones — and what none of them can do is work inside your customer's portal under supervision and remember what your team decided. Bring whichever you are leaning towards to the first call and we will connect it while we talk. The point is to do in minutes what your team currently spends hours on, not to make you build something first.

How long until we see something real?

One call to check your portal is one we handle, then about fifteen minutes to connect your agent — the steps are in the docs, so you can read them before you talk to us. The first questionnaire is the demo.

What happens to an answer the agent is unsure about?

It waits for you, with the reason it stopped. Unsure answers are never entered, and an answer already in the portal is never overwritten by the agent.

Can two of us work on one questionnaire?

Not today, and we would rather say so plainly. The session runs on Webfuse, which is built for exactly this — two people in one live browser, looking at the same question — but the space we ship is set to one participant at a time. Turning it on is a configuration change on our side rather than something to build, so if you need it, ask. Speak to us about it.