Security review for law-firm dictation.
A public checklist for firms that want to understand where audio, raw drafts, dictionary entries, snippets, and optional cloud services come into play before adopting speech-to-text.
Review Questions for Firms
TypeWhisper is a speech-input tool for editable drafts. Review should focus on data flow, storage, optional integrations, and manual approval boundaries.
- 1Which audio and text data stays on the device when local engines are used?
- 2When would an optional cloud engine or LLM provider be contacted?
- 3Where are history, dictionary entries, snippets, profiles, and API keys stored?
- 4Is the local API enabled, where is it bound, and how is access limited?
- 5How can the firm deliberately avoid cloud features for sensitive tests?
- 6Which professional review remains required before sending, deadlines, or client communication?
Data Flow by Building Block
Local processing
On-device engines can process speech locally. The selected engine should be reviewed deliberately before use.
Optional cloud
Cloud providers and LLM add-ons are not required for dictation. If used, they should be configured deliberately and approved internally.
Local storage
History, dictionary entries, snippets, and profiles should be reviewed separately because they can contain different content and risks.
API & automation
The local API is an optional power-user building block. Firm IT should decide whether it is needed and how access is constrained.
Professional approval
TypeWhisper creates raw drafts. Legal merits, deadlines, citations, tone, and sending approval remain with the firm.
Deliberate Boundaries
- No blanket promise that professional duties, client confidentiality, or privacy obligations are automatically solved.
- No legal advice, deadline control, or automatic professional documentation.
- Do not use real client or customer data in demos before internal approval and data flow are clear.
Questions about the security review?
Send us your review questions — we answer with what we can document publicly and point you to source code and docs where it helps.