Local
For engineering and security teams. What runs on your hardware, what crosses the line, and what leaves the building · which is nothing. It shows up when the tools get out of the way and let people think.
[01 / THE PROBLEM]
Conversation analysis normally runs on somebody else's servers.
Audio, transcripts and scores leave the building to be read. In healthcare, finance, government contracting and outsourcing where the data has to stay in the country, that is not a preference. It is the thing the contract forbids.
[02 / THE SESSION]
A technical session, run with the people who will sign the architecture diagram.
The session covers three things: the models, the pipeline and the console. What each one does, and where each one runs.
01 · MODELS
Installed once, no outbound inference.
Speech, reasoning and scoring models run on your hardware. There is no key to a public API because the transcript never crosses to one.
02 · PIPELINE
Ingest, redaction, analysis, signing · all inside.
Transcription, personal-data removal and the eight modules run as one local pipeline. The audit chain is written to your storage.
03 · CONSOLE
Operator, auditor and executive · served from your network.
Your identity provider, your access rules, your retention. Same build as the hosted console, installed where security can point at it.
[03 / BOUNDARY]
What crosses the line, and what never does.
The session is structured around the questions your security review actually asks · not a marketing claim about on-premise.
-
01 · AUDIO · TRANSCRIPT
Stay. Recordings are read where they are stored. Redaction happens before any module reads the text.
-
02 · MODEL WEIGHTS · RESULTS
Stay. Weights do not phone anywhere to run. Scores and the signed chain are written to your database.
-
03 · UPDATES
Come in. Signed release bundles you install on your schedule, and the licence check. Nothing else.
-
04 · SUPPORT
Runs from logs you can read first, in a session you open, with an audit entry either side of it.
Every line above is a question your security review will ask. The session answers them against your environment.
[04 / OUTPUT]
You leave with the diagram, drawn against your own environment.
The session ends with the architecture on paper and every boundary question answered on it, agreed with the people who sign it. Not a proposal to read afterwards.
The diagram
MODELS · PIPELINE · CONSOLE
What each one does, and where each one runs inside your perimeter.
The answers
ONE PER BOUNDARY LINE
What stays, what comes in, and what support can see. Answered against your environment, not against a marketing claim about on-premise.
The sizing
YOUR QUESTION, WORKED THROUGH
The sizing question you brought, taken against the systems that sit inside the perimeter.
0
INTERACTION TEXT OUT
Nothing leaves the building on inference. That is the point of the session.
Deployment is scoped separately against your environment. The full REVEAL Local page and the model table are published, if you want the detail before we meet.
[05 / WHAT TO BRING]
Your sizing question and your non-negotiables.
In person or remote. The length is agreed on the call, because it depends on how many systems sit inside the perimeter.
If the scorecard should come first, that session is the scorecard, end to end.
[06 / BOOK]
Book this session.Bring your sizing question.
Thirty minutes to decide whether a technical deep-dive is the right next step, or whether a working session on your scorecard would come first.
quality visible.