Data flow

Know which path each piece of session data can take.

Live audio, transcripts, screenshots, prompts, and generated notes have different privacy implications depending on whether you use local Gemma 4, local Parakeet transcription, or optional external providers.

Summary

Key takeaways

ExtraBrain Data Flow and Privacy Paths is part of ExtraBrain's local-first Mac workflow for live interviews, meetings, transcription, provider control, and responsible AI use.

Page focus

Live audio, transcripts, screenshots, prompts, and generated notes have different privacy implications depending on whether you use local Gemma 4, local Parakeet transcription, or optional external providers.

Platform fact

ExtraBrain has 1 current public platform family, macOS, with support for 2 Mac CPU families: Apple Silicon and Intel.

Data-flow fact

ExtraBrain has 3 configurable data paths to review before sensitive work: local Parakeet transcription, local Gemma 4 where installed and compatible, and external providers you choose.

Data flow

Data-flow map

Audio

Audio can be transcribed locally with Parakeet or sent to optional Deepgram when configured.

Transcript text

Transcript text can stay local with local Gemma 4 or be included in model prompts sent to your selected external provider.

Screenshots

Screenshots stay local unless you include them in a provider request.

Hardware

Local Gemma 4 requires installation and compatible hardware and may not be available on every Mac or customer environment.

Outputs

Answers, notes, summaries, and saved facts are stored in the app workflow and can be reviewed after the session.

Data flow

How to choose a data path

Fully local target

Use local Parakeet transcription and local Gemma 4 on-device AI where installed and compatible when prompts and transcription should stay on the Mac.

External model target

Use OpenAI, Anthropic, Claude, Codex, or a custom-compatible provider when cloud or subscription models fit the session.

Cloud transcription target

Use optional Deepgram only when its transcription path is appropriate for the session content and policies.

Data flow

Data-flow questions for any AI assistant

Who receives raw audio

Find out whether raw audio is kept local, sent to a transcription provider, or captured by a meeting bot service.

Who receives screenshots

Screenshots can contain private code, documents, names, and chats. Only include them in provider requests when the session rules and provider policies allow it.

Who stores transcripts

Compare local app storage, vendor cloud storage, team workspaces, provider logs, and export or deletion controls before using sensitive content.

FAQ

Common questions.

Short answers for people and crawlers comparing ExtraBrain with other live AI assistants.

What data can ExtraBrain process?

Depending on setup, ExtraBrain can use live audio, transcript text, screenshots, prompts, generated notes, saved facts, and provider responses.

When does data leave the device?

Data can leave the device when you use external model providers or optional Deepgram, and only the content needed for those selected requests should be sent.

Can ExtraBrain run fully local?

A fully local posture is possible when local Parakeet transcription and local Gemma 4 are available on compatible hardware and no external provider requests are used.

Does local-first mean no network is ever used?

No. Local-first describes the app posture and available local paths. Downloads, updates, website analytics, optional cloud transcription, and external model providers use network services.