Some companies will not let an unfiled invention leave the building. Fair enough. For them, eety runs on-premise; the application, the workers, the database and every file sit inside your network. Only the AI model and the search engines are outside... and this page tells you exactly what goes to them.
It is the same application either way: one record per invention, with harvesting, prior-art search, drafting, review and outside counsel all on one matter. Read the one-page fact sheet.
Tier one
Hosted by eety on Google Cloud. You sign up today, start on free tokens and pay per service. Underneath sit tenant keys on every customer table, a per-matter ledger and an audit log of every action.
For individual practitioners, firms and companies that are comfortable with a well-run cloud. Start free or see the prices.
Tier two
The application, the workers, the database and every file live inside your network, on your servers or your private cloud. eety engineers deploy and upgrade it under a services agreement.
For companies whose policy says "internally hosted only". Three things leave, and we list them below; nothing is sent to eety.
Think of the deployment as a building with three doors. Everyone and everything that matters lives inside: your people, your records, your files. Three things go out through those doors, each for one job... and nothing goes out through any other. So why spell it out this carefully? Because "your data never leaves your network" would not be true of any AI product, ours included.
Door one
Google Gemini, via Vertex AI
Through this door go the text, documents, images and audio of each AI step: a disclosure being understood, a section being drafted, a counsel draft being reviewed. Generated drawings come back from Google's image model. All of it runs under Google Cloud terms, in a Google Cloud project you can own.
Door two
Patent and literature databases, via SerpApi
Search strings and publication numbers, nothing more, during the state-of-the-art search in harvesting and during a prior-art search. Never your disclosure, never your draft, and never a file of any kind.
Door three
Public web pages
The server fetches these pages itself, during the literature and known-art steps of a prior-art search. The page's publisher sees your server's address and the URL, the way any website sees a visitor. You can close this door, at the cost of those two steps.
Want the exact list of what goes through each door, with Google's own statements quoted? It is on the one-page fact sheet.
Here is every piece of an on-premise deployment: where it runs, inside your network or outside, and whether you need it at all. Hand this table to the person who writes your firewall rules; it was written for them more than for you :)
| Component | Where it runs | Required or optional |
|---|---|---|
| Reverse proxy and TLSnginx, your certificate | Inside your network | Required |
| Web applicationReact, served as static files | Inside your network | Required |
| APIFastAPI, in a Docker container | Inside your network | Required |
| Background workers8 today, 3 high-priority and 5 low; the count is the main dial | Inside your network | Required |
| Connection poolerbetween the application and the database | Inside your network | Required |
| PostgreSQLwith the pgvector extension; matters, drafts, ledger, audit trail | Inside your network, your database | Required |
| File storedisclosures, drawings, reports, exports | Inside your network, local disk; or a bucket you own | Required, one of the two |
| Document processingOCR, Office conversion, Word, Excel and PDF generation | Inside your network, in the application containers | Required |
| Admin consoletoken grants, user suspension, usage reports | Inside your network, separate container | Optional |
| Backupsdatabase dumps and the file store | Inside your network, run by you | Required, yours |
| Mail relayinvitations, password resets, notifications | Your relay, or an external one | Optional |
| Google Geminithrough Vertex AI; text, image and audio understanding, drafting, drawings | External: Google Cloud, in your project or eety's | Required for every AI feature |
| SerpApiGoogle Patents, Google Scholar and the open web | External: SerpApi | Required for the state-of-the-art search, the density verdict and prior-art search; not needed for drafting and review alone |
| Public web page fetchesknown-art URLs and literature hits | External: public sites, patent offices, Google Patents | Required for the literature and known-art steps; can be switched off |
| Google sign-inalongside email and password | External: Google Identity | Optional |
| Card paymentsbuying token packs in the app | External: payment provider | Optional; not needed on-premise, where tokens are granted by your admin |
We could print a minimum spec here, the way most product pages do. Actually, let me back up; we have never tested a minimum, so we describe the footprint we actually run today instead. It works like the counters at a bank: a quiet morning needs one or two open, and a lunchtime queue needs more. So your sizing follows from how many searches and drafts you run at once, and the worker count is the dial.
The host
4 vCPU, 16 GB
That host runs the application and 8 workers today, with the database on a separate server. Linux, with Docker. You can put PostgreSQL and the reverse proxy on the same machine or give them their own; both work.
Outbound, fixed
An allow-list
The fixed endpoints are few, and we name them: the Vertex AI endpoint for Gemini, SerpApi, your mail relay, and Google's sign-in service if you use it. Your firewall can allow those by name and nothing else.
Outbound, open
HTTPS, or a proxy, or off
The literature and known-art steps of a prior-art search fetch public pages wherever the search results lead, and you cannot list those places in advance. So they need general HTTPS egress or an outbound proxy. Switch them off and the search keeps its patent steps but loses those two.
There is no self-serve installer. We would rather tell you that now than hand you a download that does not work... and if you have run on-premise software before, you have probably been handed one of those at least once. eety engineers install the deployment; configure it for your Google Cloud project, your mail relay and your egress rules; and upgrade it on a schedule agreed with you, all under a services agreement.
Backups are yours. That follows from the design: the database and the file store are on your servers, so your existing backup practice already covers them. We tell you what to include, and we help you test a restore before go-live.
Installation
Done by eety engineers, on your host or your private cloud, with your network team in the room for the proxy, the certificate and the firewall rules.
Upgrades
Scheduled with you, carried out by eety engineers. Nothing updates itself, and no upgrade happens without your window.
Backups and restore
Yours to run: PostgreSQL dumps and the file store. We document what to back up and sit with you for a restore test.
Administration
Your admin grants tokens, suspends users and reads usage in the admin console, inside your network. The per-matter ledger and audit log are the same as in the cloud.
Every AI step in eety is Google Gemini, called through Vertex AI. The model does not run inside your network, and we will not pretend otherwise. What you do choose is which Google Cloud project makes the call. Think of a courier: the van and the driver are the same whichever account a shipment is booked on, but the account decides whose name is on the contract, whose limit it counts against, who gets the invoice and who keeps the tracking history. So the project you pick decides who holds the terms, the quota, the bill and the logs.
Variant A
In your own Google Cloud project
The deployment calls Gemini with a service account you issue, in a project you own. Google bills you directly, the request logs are in your account, and the Google Cloud terms are between you and Google. eety never sees a prompt.
You provision the project and enable Vertex AI; we configure the deployment against it.
Variant B
In eety's Google Cloud project
The deployment calls Gemini through eety's project, as Eety Cloud does. Same model, same Vertex AI endpoint, and the same Google Cloud terms. Usage is attributed to your deployment and shown in your ledger; the Google bill is ours.
The quicker start if you do not have a Google Cloud account handy.
Not used to train
In both variants Gemini is used under Google Cloud terms, which say Google will not use your data to train or fine-tune models without your permission. Google's retention and abuse-monitoring settings are Google's, and the fact sheet quotes them.
Region
The endpoint is a Vertex AI region where Gemini is available, set per deployment and written into the services agreement. We do not name a country on this page, because "data residency in X" is only as true as the region configured and the model's availability there.
What goes to the model
Only what the task needs, per request: the disclosure text, the documents and images attached to it, audio for transcription, the section being drafted or reviewed. Drawings are generated by Google's image model, so figure content travels both ways.
What we do say
Runs on your infrastructure. Your documents, drafts, reports, search results and audit trail stay in your database and your file store. Document processing happens inside the deployment. Nothing is sent to eety: no telemetry, no license ping, no usage upload.
What we do not say
"Air-gapped", "your data never leaves your network", "private AI model", "SOC 2 certified", "one-click install", "automatic updates", "built-in backups". None of those is true today, so you will not find any of them on this page.
What you can ask for
A DPA, the sub-processor list, a completed security questionnaire, and the full "what leaves your network" sheet. Deployed to enterprise customers by eety engineers.
No, and we will not write that sentence. Three things leave: the text, documents, images and audio of each AI step go to Google Gemini through Vertex AI; search strings and publication numbers go to Google Patents, Google Scholar and the open web through SerpApi; and the literature steps of a prior-art search fetch public web pages directly. Everything else, including the database, every file, every draft and report, the ledger and the audit trail, stays inside your network. Nothing is sent to eety.
No. eety runs on Google Gemini, called through Vertex AI. There is no private model and no bring-your-own-model option. What you can choose is the Google Cloud project that calls Gemini: your own, so the terms, the quota, the bill and the logs are yours, or eety's.
Either yours or eety's; the deployment is configured one way or the other. In both cases Gemini is used under Google Cloud terms, which exclude training on your prompts. The endpoint is a Vertex AI region where Gemini is available, set per deployment. We do not promise a named country on this page; the region is agreed with you and written into the services agreement.
Yes. The server fetches public pages only during the literature and known-art steps of a prior-art search. With those fetches off, the search still runs its patent steps through SerpApi, but the literature and known-art steps are lost. Search strings and publication numbers still go to SerpApi. Prior-art search itself is enabled per organization on request.
eety engineers deploy and upgrade it under a services agreement, on a schedule agreed with you. There is no self-serve installer and no automatic update. Backups of the database and the file store are yours to run; we tell you what to back up and help you test a restore.
Not today. There is no SOC 2 or ISO 27001 certificate, no SAML or OIDC single sign-on, no SCIM provisioning and no enforced MFA. Sign-in is email and password or Google sign-in. We list the controls we actually run on the fact sheet, and we answer security questionnaires on request.
We walk through the three doors, the component table and the Google Cloud project choice with the people who will write the firewall rules; then we show the product on a matter like yours. The one-page fact sheet is yours to circulate beforehand. I keep looking for a shorter way to explain all of this... maybe on-premise is simply not a short subject.