transcribe-assistant is an offline, low-latency speech-to-text engine.
Security and privacy are core design principles:

• no cloud APIs
• no external telemetry
• local-only processing
• audit-ready sessions (PRO roadmap)

• All audio is processed locally on the user’s machine.
• No audio, transcripts, or metadata are sent to:
  • OpenAI
  • Google Cloud
  • Azure
  • any third-party API or service.
• No tracking, analytics, or telemetry are embedded in the engine.

Conclusion:
All data stays on the device where transcribe-assistant is running.

• The engine uses Faster-Whisper as a local ASR backend.
• Models are loaded from disk and executed locally.
• No remote model calls are performed.

Users are encouraged to:

• review dependencies in requirements.txt
• pin versions for reproducibility

• Transcription sessions may be stored locally as:
  • text files (TXT)
  • structured data (JSON)
• Storage location is configurable.
• Users are responsible for:
  • filesystem-level access control
  • encryption at rest (if required)
  • backup policies

By default:

• transcribe-assistant does not require internet access.
• no outbound connections are made by the core engine.

If you integrate transcribe-assistant into a larger system (API, SaaS):

• you must document your own network and security model.
• this repo does not include any cloud integration by default.

transcribe-assistant is designed for:

• offline environments
• sensitive audio (NGO, legal, medical, journalism)
• systems where cloud-based ASR is not acceptable

Primary protections:

• no external data exfiltration
• local-only processing
• transparent, inspectable open-source code

The planned deterministic PRO engine will add:

• reproducible pipelines
• deterministic timestamp alignment
• audit-ready logs
• compliance-focused workflows

These features are intended for:

• regulated environments
• forensics
• legal and compliance use cases

If you discover a security issue:

• please contact: krunixbase@gmail.com
• include:
  • description of the issue
  • steps to reproduce
  • potential impact

We aim to:

• acknowledge reports promptly
• provide fixes in a reasonable timeframe
• document security-relevant changes in release notes

We recommend:

• running transcribe-assistant on trusted machines
• using up-to-date OS and security patches
• restricting access to stored transcripts
• using disk encryption where appropriate

transcribe-assistant is provided “as is”, under the MIT License.
Security in production environments depends on:

• your infrastructure
• your OS configuration
• your access control and policies

Always review and adapt this tool to your specific threat model.