audio transcription

Khoj maintains 100% search uptime through a deterministic offline fallback and local OpenAI Whisper transcription for private data processing.

Khoj indexes your local files (Markdown, PDF, and Org-mode) to provide a private, self-hosted AI assistant. It uses the OpenAI Whisper model to handle audio transcription directly on your CPU or GPU: this keeps your voice data off the cloud. When network access fails, the system activates a deterministic dummy embedding fallback (mapping new data to fixed vectors) to keep your search index operational. This architecture ensures you never lose access to your RAG (Retrieval-Augmented Generation) workflow, even in zero-connectivity environments.

https://khoj.dev