Grounded drafts
Structured prompts use the meeting record and defined output shapes. These safeguards aim to reduce unsupported content; they cannot guarantee an error-free result.
Product specifications
The details
behind the experience.
The local meeting workflow, the hardware and the engineering choices behind it. Confirm the supplied configuration and release-specific evidence as part of your evaluation.
Core product facts.
- Product type
- Dedicated private-AI meeting appliance, built around NVIDIA Jetson Orin, NVMe storage and a considered enclosure with active cooling. Confirm the supplied hardware configuration.
- Core workflow
- Capture, on-device transcription and speaker tooling, review, structured notes, a session snapshot, Theia and deliberate export.
- Connectivity and accounts
- No Wi-Fi hardware in production devices. No cloud account or meeting bot required for the local recording workflow.
- Telemetry
- No appliance telemetry: meeting content, transcripts, prompts, usage data and device activity are not sent to Noter. Deliberate exports and website enquiries are separate.
- Transcription
- On-device speech recognition with custom diarisation and speaker tooling. Meeting context, names and terminology support recognition; review remains necessary.
- Theia
- A local language model tuned for meeting work through prompt engineering, context-window and temperature calibration, and response-shape testing. It works with the selected meeting record.
- Cloud upload
- Not required for transcription, structured notes or Theia.
- Device vault
- AES-256-GCM encryption with configurable password protection. Vault keys use PBKDF2-HMAC-SHA256 at 240,000 iterations.
- Vault coverage
- Session metadata, transcript segments, structured note JSON, session snapshots, Theia conversation messages, the encrypted index and retained audio. Verify coverage and configuration for the supplied release.
- Storage capacity
- Published capacity: up to 7,000 hours of raw audio or up to 1,000,000 completed notes with transcripts, whichever fills first. These are not simultaneous capacities; actual capacity depends on hardware, record size and retention settings. Ask for the relevant test conditions.
- Retention
- Raw-audio retention can be managed separately from the transcript and note. Agree access, deletion, recovery and export handling for the evaluated release.
- Export formats
- User-approved PDF, DOCX and TXT. Exported files are deliberate copies outside the encrypted device vault.
- Output review
- Check transcription, speaker attribution, names, decisions, actions and AI answers before relying on or sharing them.
Under the surface
Considered at every stage.
Session-focused Theia
The selected transcript, note and snapshot provide the working context. Answers remain subject to checking against the source.
Transcript hygiene
Noise handling and duplicate filtering support cleaner text before note generation. Evaluate performance in the rooms and audio conditions you expect.
Context before capture
An agenda, glossary, speaker information and key terms help the local workflow interpret the discussion.
Separate capture and inference
Recording and AI processing use separate execution paths with buffering, designed to keep capture responsive while inference runs.
Fresh processing state
Each recording starts with fresh processing state. This is distinct from retention: intentionally saved records remain governed by the vault and your settings.
Local speaker profiles
Voice-profile enrolment and speaker tooling run on the appliance. Decide when enrolment is appropriate and verify attribution.
Protected session records
Authenticated vault encryption and configurable password protection support the local storage boundary. Device access and exported copies need their own controls.
Specify the environment.
Then measure the fit.
Discuss microphone and room setup, connectivity, power, physical dimensions, supported languages, storage, support and updates for the unit being evaluated.
Capacity and processing speed depend on configuration and workload. Ask for test conditions and release-specific results rather than relying on an unqualified headline number.
Discuss your deployment