Data transfer
A cloud service processes the recording in the provider's infrastructure. Before use, review the storage region, retention period, subprocessors, and access controls.
SUB–02 / MEETING · Local Meeting Scribe · Runs on your computer
One audio recording of a meeting — four working documents. Locally*.
You upload a recording of a meeting or interview, or a document. The Scribe transcribes the speech, detects the speakers, translates if needed and assembles the set: a summary, minutes, a report on your request and a task list. Verified against the source recording. In local mode, data stays within the selected environment*.
* Local mode is the primary one and is on by default. External services are connected only on the user's explicit instruction and with their own access keys.
01Problem
When local deployment may be preferable.
Four factors to review before sending meeting recordings to a cloud service.
Confidentiality terms
Recordings may contain trade secrets, privileged information, medical information, or personnel decisions. Cloud processing requires a review of the provider agreement, data-handling terms, and internal policies.
Threat model
Cloud processing adds account, API, and provider-infrastructure risks. Review authentication, encryption, retention, deletion, and incident-response controls.
Pricing model
Cloud services may charge by audio minute, usage volume, subscription, or tokens. Estimate recurring costs and limits before processing meetings regularly.
A local configuration with no external integrations avoids sending the recording to a cloud provider: materials remain in the approved environment, and every document is checked against the source recording.
02Route
Six stages — from uploading the file to a verified document set.
Every stage runs locally* and is recorded in the event log.
File upload
Audio WAV, MP3, M4A, FLAC, OGG — or a document TXT, Markdown, PDF, DOCX, RTF. In local mode, the file stays within the selected environment.
Transcription
Speech is turned into text on your hardware. The recognition engine is chosen for the task.
Diarization
Who said what and when — with a timeline for each participant.
Translation
A foreign-language recording or document is automatically translated into Russian.
Report preparation
Summary → minutes → report on request → task list. A strict sequence: the task list absorbs the action items of all previous stages.
Document export
The finished set is exported in DOCX, PDF, Markdown, TXT, CSV, XLSX, JSON; additionally — export to Obsidian with metadata and internal links.
* The user chooses whether data is processed locally or in the cloud. The agent works both locally and in the cloud.
03Documents
Four documents from one source.
They are generated in a strict sequence — that is why the task list includes the action items of all previous stages.
Summary
A condensed account of the key outcomes of the discussion.
Who needs it: The manager: the essence of the meeting without reading the full transcript.
Minutes
Participants, agenda, the course of the discussion, decisions made and action items.
Who needs it: The team: an official meeting document in a single format.
Report tailored to the user's request
The user can change the prompt for generating the report. The document is prepared for your request: only financial decisions, only technical questions — whatever you need.
Task list
All action items and activities, consolidated from the results of the previous stages.
Who needs it: The assignees: not a single action item is lost.
The order is not accidental: summary → minutes → report on request → task list. Each next document builds on the previous ones, so the task list captures every identified action item — none is lost between stages.
04Local
In local mode, data remains in the approved environment.
Core stages — from file intake to report export — run on your hardware*; external connections depend on the selected configuration.
- In a configuration with no external integrations, recordings, transcripts, and reports are not sent to external servers
- Local mode is the primary one and is on by default
- External services — only on explicit instruction and with your own access keys
- Every stage is recorded in the event log: input, processing, the basis of each document
- A complete evidence base: on which content each report was generated
| Parameter | Cloud transcriber | Local Meeting Scribe |
|---|---|---|
| Where the recording is processed | On the provider's server | On your computer |
| Internet for operation | Depends on the service | Not required for core local functions; external integrations may require it |
| Access to meeting content | Depends on the provider agreement, settings, and processes | Limited by configured roles in the local environment |
| Use of data for training | Depends on the provider terms | The local configuration does not transmit data for external training |
| Confidential meetings | Require a contract and risk-model review | Processed under the access settings of the approved environment |
| Result | Depends on the selected service and plan | 4 documents checked against the source recording |
* The user chooses whether data is processed locally or in the cloud. The agent works both locally and in the cloud.
05Result
Export to where the document will live.
Seven formats, export to Obsidian and an event log — the report fits into your workflow.
7 export formats
From Word to JSON
DOCXPDFMarkdownTXTCSVXLSXJSON
The processing results are exported in formats for people and for systems — without manual conversion.
Integration
Obsidian
Report documents are exported to the vault in Markdown format with automatic metadata population: task identifier, date, source, document type — and with internal links generated.
Evidence base
Event log
What data came in, which stages were completed, on the basis of which content each document was generated. Segments transcribed with low confidence are marked for human review.
06Audience
Who the Scribe saves hours for every week.
Each role gets its own document from one set.
Project manager
Minutes with decisions, action items and assignees — right after the meeting, in a single format.
A participant who missed the meeting
A one-page summary: what was decided, what was assigned, which responsibilities belong to them.
Analyst, researcher
A transcript with a speaker timeline and a report on a given topic from a series of interviews.
Finance and technical teams
A custom report: only financial decisions or only technical questions — on request.
Lawyer, doctor, auditor
For private meetings, the recording and reports are processed within the approved local environment.
A team with a knowledge base in Obsidian
Reports land directly in the vault — with metadata and internal links.
07Scenarios
Where one recording becomes a document set.
Typical situations: from a weekly status to negotiations in a foreign language.
🗓️ Weekly project status
A meeting of 12 people, 90 minutes, three parallel workstreams.
What is generated
- Minutes: agenda, decisions, action items
- A one-page summary for management
- Task list: 14 action items with assignees
What it gives
- Automated preparation may reduce the time required compared with manual transcription
- Action items are consolidated into one list for review by the meeting owner
🎯 A series of client interviews
6 interviews of 40 minutes each, different speakers and speaking styles.
What is generated
- Transcripts with a speaker timeline
- Report on request: only pains and objections
- A consolidated list of topics for research
What it gives
- Materials for analysis ready on the day of recording
- Participant quotes — with exact timecodes
🌐 A consultation in English
Negotiations with a foreign contractor, a recording and a PDF contract.
What is generated
- Translation of the transcript into Russian
- Minutes with the commitments recorded
- Export to DOCX and Markdown for Obsidian
What it gives
- The content of the negotiations is clear without an interpreter
- The agreements are recorded and verified against the recording
Boundaries
What the human checks.
Quality depends on the recording, the speech, the terminology and the number of participants. The names of the speakers, the decisions made and the deadlines of action items are confirmed by the person responsible for the meeting. The Scribe prepares the document — the decision stays with the human.
What the Scribe does not do
- Does not make decisions for the meeting participants
- Does not send or publish documents without confirmation
- Does not replace a court reporter in legal proceedings
- Does not work with recordings whose processing the participants have not consented to
08Questions
Short answers to 10 frequent questions.
Direct answers for quick reading. Expand the question you need or open all at once.
What does the client get at the end?
Four documents from one recording: summary, minutes, custom report and task list. Plus a full transcript with a speaker timeline.
How does local mode work?
In local mode, file intake, transcription, analysis, and export run on your hardware. Recordings, transcripts, and reports stay within the selected environment.
Is the internet required?
Not for the core functions. External services are connected only on your explicit instruction and only with your own access keys.
Who can see the meeting content?
In local mode, only authorized users within your environment can access the content. Local mode is on by default; connecting external services requires the user's explicit choice.
Which formats does it accept?
Audio: WAV, MP3, M4A, FLAC, OGG. Documents: TXT, Markdown, PDF, DOCX, RTF. The list of supported formats can be extended at the customer's request for an additional fee.
Where are the results exported?
DOCX, PDF, Markdown, TXT, CSV, XLSX, JSON. Separately — export to Obsidian with metadata and internal links.
How is the reliability of reports verified?
Each report is automatically checked against the source recording: the system confirms that the conclusions are based on the meeting content. Every stage is recorded in the event log.
What does the human confirm?
Participants, the wording of decisions, deadlines and assignees. Segments transcribed with low confidence are marked for review.
What hardware is needed?
Windows 10/11, at least 16 GB of RAM and an NVIDIA graphics card with at least 6 GB of video memory. The more powerful the computer, the faster the processing.
How is this different from a cloud transcriber?
A cloud service processes the recording in the provider's infrastructure; storage and access terms depend on the service. The Scribe produces four documents and, in local mode, processes the recording within the approved environment.
First step
Choose a ready-made solution or describe your process.
Pick one of the twelve agents and subagents, or fill in the questionnaire for custom development.