Bridge AI

Data Trust and Compliance Commitment

Effective 18 September 2026

Your accounts remain yours. Bridge AI only accesses what is required to operate the service.

We cannot read your Gmail inbox or your sent mail. The grant we ask for cannot read mail at all.
We never use your data to train AI models, and every model call refuses providers that would.
Nothing is sent from your accounts without your approval.
Call recording only starts when you start it, and the audio never reaches our servers.
Your workspace is isolated from every other expert's, enforced on the server.
You can disconnect any integration, or ask for deletion, at any time.

Exactly what Bridge AI can access

Five connections, and this is all of them. A grant we did not ask for is a grant we do not have.

Tool	Status	Reads	Writes	Where it lives
LinkedIn	Required	The expert's LinkedIn inbox and the replies to their campaigns.	Only the messages the expert approved, sent from their own account.	Synced to Bridge AI.
Gmail	Required	Nothing. The inbox and the sent mail are never opened.	The emails the expert approved, from their address, into their own Sent folder.	Synced to Bridge AI: the access token, nothing else.
Google Calendar	Required	Events, to compute availability and to log the meetings held.	The discovery and debrief events booked through the expert's page.	Synced to Bridge AI.
Granola	Optional	The meeting notes and transcripts the supplied API key gives access to.	Nothing.	Synced to Bridge AI, so a brief can quote the call it came from.
Bridge AI Recorder	Optional	The audio of the calls the expert chooses to record, and nothing they do not start.	Nothing.	The audio never reaches Bridge AI. It goes from the expert's Mac to Deepgram for transcription, under a four-hour key. Only the text comes back to us.

LinkedIn is required, not optional: campaigns run on LinkedIn, so an expert without it has nothing to send and no reply to track. Granola and the recorder are genuinely optional. The system works without either, it simply has fewer calls to quote from.

What is never read

Gmail inbox and sent mail. The Gmail grant cannot read mail.
Files on the expert's computer, other than the recordings they start themselves.
Any other expert's workspace. Every query is scoped to one workspace and enforced on the server.
Anything at all, by us, for model training.

Where your data lives

ApplicationVercel. The functions that serve the workspace are pinned to Singapore.

DatabaseNeon managed Postgres, on AWS in Singapore (ap-southeast-1). One database, with every query scoped to a single workspace.

In transitTLS on every connection, to the workspace and to every provider below.

At restThe managed Postgres is encrypted at rest by the provider. On top of that, every access token and API key we hold is encrypted again by us, with AES-256-GCM, before it is written. Tokens never reach a browser and never reach a log.

Call audioThe audio captured by the Bridge AI Recorder never reaches Bridge AI. It is streamed from the expert's Mac to Deepgram for transcription, under a key that this system mints for that session and that expires in four hours. Only the text transcript comes back and is stored. The recorder's device key stays on the Mac.

AI processingModel calls go through OpenRouter to Anthropic (Claude) and Google (Gemini), and carry the working context for that one request and nothing else. Bridge AI trains no models of its own, and every call refuses to be routed to a provider that stores or trains on what it is sent: the request itself carries that instruction, so it is not a setting somebody can change without changing the code.

Other processorsUnipile (the LinkedIn connection), Apify (public LinkedIn profile data), DataForSEO (web search for lead research), Deepgram (transcription, above), Voyage AI (the embeddings that make a workspace searchable), fal (images generated for the expert's own content), Granola (transcripts, optional), PostHog (how the workspace itself is used) and Google's own APIs for Gmail sending and Calendar. That is the complete list. Each receives only the request it serves, under its own terms.

Two kinds of data, and we hold one of them

Expert workspace data is how the expert wins work: campaign activity, LinkedIn conversations, calendar availability, meeting transcripts, the notes and documents they upload about their own positioning. That is what this page is about, and it is what the system holds.

Client engagement data is what an expert sees once a client has hired them: financial statements, HR records, internal documents, management reporting. Bridge AI does not ask for it and is not built to hold it. There is no client file, no data room and no place in the product where an engagement's documents belong.

An expert can upload a document into their own workspace, and some do, so the honest answer is that anything they upload is stored under their workspace and under everything on this page. If a client's material must stay out of it, that is a rule between the client and the expert, and it is the expert who keeps it.

Who can access it, and how access ends

The expert, and the named Bridge AI operator who runs their service. Nobody else. That operator access exists because the service is run for the expert rather than sold as self-service software, and every operator action is recorded against a named account.

Disconnecting a tool in the workspace revokes its token immediately and stops the system using what it read through it. Deletion is one policy, not two: ask at any time, including when an engagement ends, and a full export and deletion of the workspace completes within thirty days. Write to leonard@bridgeaimena.com.

Twelve commitments

01We take the minimumOnly the data the work needs, kept only while it serves that purpose. One deletion policy, stated above, applies to all of it.
02Confidentiality by agreementEvery engagement runs under a confidentiality agreement, and it survives the engagement. What we hold and what we will never ask you for are written into the Terms, not only here.
03A named, short list of providersThe providers in the chain are named on this page rather than described. Each runs under its own data processing terms, and we will show you which on request.
04Identity removed where it is not neededWhere the work does not need to know who someone is, the identifying fields are reduced or replaced.
05Transparency and the right to askAsk what we hold, why, where it sits and for how long, and you get an answer in plain language, not a policy extract.
06Your data stays yoursYou own what you put in. We hold it and process it on your behalf, and we claim nothing else.
07Audit and accountabilityYou can ask for our security practices, the providers in the chain, and the record of operator actions taken on your workspace. Every one is written against a named account.
08Processed where we sayThe application and the database both run in Singapore. The providers named above process elsewhere, each under its own terms, and none of them receives more than the request it serves.
09Security by designTLS in transit, and a managed Postgres that is encrypted at rest. Tokens and API keys are encrypted again by us before they are written. Every query is scoped to one workspace on the server; there is no path from a browser to another workspace's data.
10You hear about a breach in 24 hoursWe investigate and contain immediately, and you hear from us within 24 hours of confirming a breach affecting your data. It is section 8 of the Terms, so it is a term and not a promise on a page.
11People and trainingEveryone with access is briefed on confidentiality and on the rules that apply, and that briefing is refreshed as the rules change.
12A human decidesAutomation drafts. A person approves. No message leaves your account and no client record changes without that approval.

Questions a client usually asks

Can Bridge AI read my emails to the expert? No. The Gmail grant cannot read mail.
Is our call recorded without us knowing? No. Recording is started by the expert, and the platform they record on shows it.
Does our conversation train an AI model? No. Bridge AI trains no models, and every call we make refuses to be routed to a provider that would store or train on it.
Who at Bridge AI can access the expert's workspace? The expert, and the named operator who runs their service. Every operator action is recorded against that account.
Do you hold our documents? No. There is no client file in the product, and section 7 of the Terms says so. Anything the expert uploads sits in their own workspace, under the rules on this page.
Where does it run? Singapore, for both the application and the database.

The full legal text sits in the privacy policy and the terms. For anything not answered here, write to leonard@bridgeaimena.com.