When you opt in, Datum sends the text you provide to our AI provider (Anthropic Claude when configured, otherwise heuristic fallbacks on-device/server). We do not send your master password or vault encryption keys.
Consent model
Each AI surface requires explicit opt-in before sending data. You can use Datum without AI — heuristic fallbacks exist for auditor extraction when AI is unavailable.
What each module sends
- Auditor (baseline): Text excerpts from locally decrypted vault files you choose to analyze. Sent only after consent checkbox.
- Signal Deck (stress, strengths, direction, skills, interview): Free-text context you paste into the module, plus optional role/company fields.
- Career tools (bio, review, roles): Context you provide. Role search returns AI-suggested titles — not live job listings.
- Runway guidance: Numbers from your calculator (cash, burn, severance, unemployment) and household size — only after consent.
- Learning mentor and coding help: The active mission or problem, the specific question you ask, your selected help mode, and up to 5,000 characters of the current practice draft. This is sent only after explicit consent for that response. If you opt in during Learning setup, a compact Datum career graph summary (phase, focus role/company, capacity, and recent coach memory) is also included. Vault files and encrypted baseline contents are not included. Practice drafts are not stored in learning progress.
What is never sent
- Master password or derived vault encryption keys
- Full vault ciphertext blobs (unless you explicitly export them yourself)
- Reel video files to the AI provider (videos are stored separately in blob storage)
- Learning drafts when you do not actively consent to an AI mentor response
Provider & fallbacks
When configured, Datum uses Anthropic Claude via API. If AI is unavailable or you decline consent, heuristics or saved local data are used instead. The UI indicates the source when applicable.