Choose what gets captured
Capture starts from an explicit allowlist. Transcription and speaker fingerprints stay local, with source controls and status information in the interface.
Conversation memory
NX Recall.
Find the conversation again.
Local transcription, persistent speaker identity, and searchable conversation history for VR and voice chat. Capture only the applications you explicitly allow.
01 — CAPABILITIESNX RECALL
Search beyond a transcript
Find conversations by speaker, date, or words. Optional local models add semantic search and a memory graph for exploring relationships and commitments.
Return with more context
Desktop captions, uncertain-row review, and later reprocessing help make speech useful after the moment has passed. Optional night-shift processing revisits the day locally.
02 — GETTING STARTED
From download to first use.
Requirements and setup vary by project. The repository documentation is the source for complete instructions.
Read the documentation ↗- 01
Install and prepare models
Install the Linux release through NX Hub or download it directly. Follow the README to fetch speech models and enable the user service.
- 02
Choose capture sources
Use the app’s source controls or recalld probe to find applications, then explicitly allow the ones you want to capture.
- 03
Review and search
Open the transcript, identify speakers, and search your history. Add semantic or graph models only when you want those optional features.
PLATFORM & SETUP
Linux x86-64 with PipeWire and a systemd user session. Desktop captions require Wayland. Model downloads need initial network access and local storage.
CURRENT SCOPE
The OpenXR headset overlay is experimental. Desktop transcription and captions should not be confused with a validated in-headset overlay.
Keep going.
Documentation, development, and help.