What leaves your machine, and what never does
Updated October 6, 2026
The short version: a quote PDF you drop to build a proposal is read in your browser and the file is never uploaded. Documents you add to an account are different. Your browser reads the text and sends it to your agency's workspace so the account can remember it. Your Vault, your saved proposals, and your brand settings are stored there too. Nothing is shared across agencies, and nothing is used for training. Two things reach an outside AI service at all, a recorded call you choose to analyze and a document you add to an account, and both are described below.
#Quote PDFs for a proposal are read in your browser
When you drop a carrier quote or a current policy in the builder, or a document on the Vault page, the reading happens entirely inside your browser. The engine that reads the PDF is bundled with the app itself, not fetched from anywhere, and runs against the raw file the moment you drop it. For a proposal there is no upload, no request to a parsing service, and no server round trip. See How ProClose reads a PDF for how the parser itself works.
That is true whether the parse succeeds or not. If a PDF cannot be read deterministically, the app falls back to the sample flow locally. The file itself is never sent anywhere to try harder. See When a PDF does not parse.
#The Vault, your proposals, and your brand reach your own server, and nowhere else
Everything in your Vault (appetite guide, fee schedule, coverage standards, vendors, and any document you drop in to update it), every proposal you save, and your Brand settings are written to your own workspace's server, so they survive a reload and follow you to a different device. That server is yours, run by whoever set up your ProClose install. It is never a third party, nothing on it is used for training, and the app never sends it anywhere else. See Where your data lives for exactly what is stored and where.
A document you drop on the Vault page to feed it new rows is still parsed locally first, the same as a quote PDF below; only the resulting appetite rows, fee lines, standards, and vendors, the structured data, get saved to your server once you confirm the merge.
#The deck renders locally
Building a proposal, editing it on the canvas or in the outline editor, and presenting it, all of that is your browser assembling and rendering the slides from data that is already on your machine. Nothing about building or viewing a deck sends its contents anywhere.
Exporting is the same story. Printing to PDF, downloading HTML or Markdown, or using Copy for AI all produce their output locally, from what is already in your browser. See Print or save as PDF and Export HTML, Markdown, and Copy for AI. Clicking Save to add a proposal to your Proposals list is different from all of that: it is a deliberate write to your workspace's server, covered above, not a local-only render.
#Documents added to an account are stored in your workspace
When you add a PDF, email, or text to an account, your browser reads the text out of a PDF and sends the text, page by page. The PDF file itself is not uploaded. The one exception is a scanned page with no text layer: your browser sends a picture of that page so it can be read, and only the recognized text is kept. ProClose keeps that text and the fields pulled from it in your agency's workspace so the account can answer questions later. Nothing is shared across agencies, and nothing is used for training. Each document is read by an AI model once, and the result is remembered. See Adding documents to an account.
#Analyzing a call
This reaches a service outside your own infrastructure, and it happens only when you deliberately pull and analyze a recorded call.
Here is the exact path:
- You pick a call in the call picker. The recording is retrieved from your RingCentral account by your own server, not your browser.
- Your server sends the call audio to a transcription service to get a transcript.
- Your server sends that transcript to an insight model to extract structured fields (prospect name, summary, coverage needs, objections, sentiment, next step, key quotes, and so on), and, in the same pass, to generate the internal coaching notes covered in Coaching notes on a call.
- The transcript, the insights, and the coaching notes come back through your server to your browser, and your server also writes them to its own disk cache so that specific recording is never transcribed, extracted, or coached a second time.
Two things are worth being precise about:
- Your browser never talks to the transcription service or the insight model directly. Both calls are made server-side. Your credentials for either service are never sent to a browser.
- This only happens for a call you choose to analyze. The Vault and a proposal you are building are never sent to the transcription or insight service. What reaches that service for a call is the call audio and the transcript derived from it.
For what gets cached, how the 25MB recording limit works, and what happens on a retry, see Call analysis: caching, limits, and privacy.
#The full picture
| Action | Leaves your machine? | Reaches a third party? |
|---|---|---|
| Drop a quote or policy PDF to build a proposal | No, read in your browser only | No |
| Add a document to an account | The extracted text, to your agency's workspace | The AI model that reads it, once per document |
| Drop a document on the Vault | Only the structured rows it finds, once you confirm the merge | No, your own server only |
| Save the Vault, a proposal, or your brand | Yes, to your own workspace's server | No, your own server only |
| Build, edit, or present a deck | No, rendered in your browser only | No |
| Print, export HTML/Markdown, Copy for AI | No, generated in your browser only | No |
| Pull and analyze a call | Yes, the call audio and its transcript | Yes, the transcription and insight-extraction services, routed through your own server |
Nothing else reaches a third party, and nothing is shared across agencies.