AI you can check for yourself.
Automate with confidence. Before a digital worker goes live, you can read its blueprint (a step-by-step plan of everything it does) and test it on made-up data. An optional module also tries known attack techniques on it, the way a hacker would. This page shows how we prove all this, and what we don’t claim yet.
EU data protection
California privacy
Encrypted when sent and stored
A person approves the key steps
Independent security audit
Top of the roadmap — target date: being finalized.
Information security
Standard for managing AI
Usable by people with disabilities
“In place” means it works today. “Roadmap” means an official certification from an independent auditor that we are working toward. We never claim a certification before we have it. For the latest status and documents, write to security@nexgile.com.
Some labels say Roadmap on purpose. We only claim what is true today, not what we hope for, and we label everything on this site the same way.
SOC 2 Type II — target date: being finalized.
Until then, here is what you can rely on:
- You run the software yourself, under the security rules you already follow.
- You can read our audit-trail specification: the list of everything that gets recorded.
- Your security team can ask to see sample files from a download, with sensitive details removed.
Security in everything we deliver.
Every download includes a threat model: a map of its security risks and how each one is handled. We run our own platform to enterprise standards.
- Data encrypted when it is sent and when it is stored
- People see only what their role allows
- A map of security risks and data flows in every download
- Continuous monitoring, activity logs and a plan for incidents
- A safe, responsible way to report security problems to us
Your data stays yours.
Workers can be tested entirely on made-up data, so none of your real data is needed to try them. After delivery, the software runs on your own systems, not on a Nexgile cloud.
- No real data needed to design or test a worker
- The software runs on your own servers or cloud
- Data residency (where your data is kept) across our Herndon, VA and Hyderabad, India operations
- Aligned with the GDPR and CCPA privacy laws. Data kept or deleted at your request
- On request: a list of the outside firms that handle data for us
AI decisions you can explain.
Keeping AI under control is part of our job, not small print. Every worker is built so its decisions can be explained and checked.
- A readable plan, not “the AI decided”
- Your rules built in wherever possible: same case, same answer
- At each approval step, the worker waits for a person
- Every approval recorded, plus an optional log of later changes. See the audit-trail specification
- Optional hiding of personal details when the worker uses its tools, off by default
Four things you can check for yourself.
Everything we promise on this site comes down to these four. You can check each one in a live session, in a document your auditors ask for, or by re-running a past test.
Rehearsal before reality
Each worker practices on made-up data shaped like its real work, before it touches anything real. You can run it step by step, watch it pause for approvals, test many workers together and even speed up time. An optional module also tries known attack techniques on it, the way a hacker would. Because no real data is needed, your security team can approve an evaluation that starts from a written description, with no data-sharing agreement.
See it in Nx·StudioRules you can rely on
The rules you give are built into the worker wherever possible, and you are told about any that can’t be. A built-in rule never uses AI, so the same case always gets the same answer. For everyday decisions that nobody has written down, there is our separate Cognitive Engine. It learns from past examples and acts only after passing a test on cases it has never seen.
Meet the Cognitive EngineBlueprints you can read
Every worker comes with a blueprint you can read: what it does, who approves what, and when a case goes to a person. Every version is saved, so you can compare changes or go back. Business teams can read it as a simple chart of who does what. Like everything the platform makes, it is yours.
See how oversight worksProof in every download
Every download includes a signed list of everything inside it (a bill of materials), and a tool your IT team can use to check it offline. It also contains the threat model and a guide that links it to the documents the EU AI Act asks for. The Studio adds a full list of what you received. The worker runs without us: no Nexgile software, no license check, no usage meter, and nothing reports back. So “no lock-in” is something you can check for yourself.
Check it yourselfDon’t take our word for any of this.
Every download comes with proof you can check yourself: a signed list of every part of the software and its AI (a bill of materials). Your IT team checks it on your own computers, offline, with no connection to us. The Studio adds a full list of what you received. You can ask for a sample, with sensitive details removed. You make the call.
The worker is built to run without us: no Nexgile software, no license check, no usage meter and nothing reporting back to us.
We treat rules the same way. In any live session, ask us to run the same case twice and watch your rules give the same answer.
- Signed bill of materials A signed list of every part of the software and its AI, plus an offline checking toolin the download
- Threat model Its security risks, how each is handled, and where data flowsin the download
- EU AI Act cross-reference Links the Act's documentation items to the bill of materials. It is not an official compliance assessmentin the download
- Ownership report A full list of everything delivered, plus a PDF summaryfrom the Studio
Checked on your own computers. You make the call.
See what a worker will do before it does it.
No black boxes. You get documents you can read, compare between versions, and keep.
One plan everyone can read
Every worker comes with a readable blueprint, and every version is saved. It shows the worker’s AI agents and their jobs, its connections, its approval steps, where each case goes and its rules. Nothing is hidden inside AI instructions or a vendor’s software. Your developers compare its versions, your compliance team reviews it before launch, and your auditors check against it later.
A record of every approval
Every approval is recorded on an audit trail: who approved what, and when. Two extras are optional and off by default: a log that shows any later change, and a record of where each piece of data came from (data lineage). Built-in rules always give the same answer to the same case.
Approvals live in your software
Approvals happen in a review inbox built into the worker itself. If you choose, requests can also go to Slack, Microsoft Teams or email. You decide who approves each step, such as a role or a team. Because it is all code you own, your approval rules are enforced in your software, not in a setting on our servers.
How we handle the data you choose to share.
Each worker gets made-up data shaped like its own work, so previews, test runs and attack tests need none of your real data. If you upload your own sample data instead, it stays in your own workspace, where only its members can see it.
Anything you share while designing is handled separately. You might share documents, process records or sample data so your workers match your real workload. You decide what to share, it is used only to design your workers, and a data-processing agreement (a contract on how your data is handled) is signed first.
If you start with a written description, that description is all you share.
What is always the same, what can vary, and who decides.
AI answers can vary from one run to the next.
That is why your rules don’t depend on AI. Wherever possible, they are built into the worker and give the same answer every time. To see this, replay any past test run step by step. Everyday decisions that nobody has written down are handled by our separate Cognitive Engine. It learns from past examples. It acts only once it passes a test on cases it has never seen.
Where AI judgment is needed, a quality check can score each answer. You set the minimum score and choose what happens below it: flag it, stop the run, or send the case another way, such as to a person. By default, it only raises a flag. At an approval step, the worker waits for a person to decide.
You decide when a case goes to a person.
Documents you can ask for.
Each one has sensitive details removed and is shared on request. To ask for any of them, write to security@nexgile.com.
A sample worker plan
The who-does-what chart and download summary of a sample worker, so your team can see what a worker will do before it does it.
A sample test report
Made from a test run on made-up data. It is the report purchasing teams ask for.
A small sample download
Its folder layout, a page of its operating guide and a test report. It shows what “ready for real work” means in practice.
An outline of the threat model
What the threat model in every download covers, without the set-up details.
The audit-trail specification
Every event the audit trail records, and how the optional log shows any later change.
A sample ownership report
The full list, the signed bill of materials and the checking tool’s results for a sample download. You make the call.
Your auditors get to see the full plan ahead of time.
Promises we can prove.
Each one is a single sentence, limited to what the platform can prove. A small promise you can check is worth more than a big one you can’t.
None of your real data is needed to design or test a worker.
Your workers run without Nexgile software. No fees to us per decision. Nothing reports back.
Nothing stops working if you stop paying us.
Every number on this site says how we measured it. If we can’t measure something honestly, we don’t publish it.
The same goes for other people’s numbers: we quote them only with their source and date. Our delivery counts say how we counted, and we re-check them every time we publish. If we can no longer confirm a figure, we remove it everywhere on the site.
Reliability
The software we deliver runs on your own servers or cloud, so you decide when it is available. Our own platform is monitored continuously, with a plan for handling incidents. A public status page for Nexgile services is planned.
Talk to the right team
- Security, or to report a problem security@nexgile.com
- Privacy & data rights privacy@nexgile.com
- Accessibility accessibility@nexgile.com