AI Transparency

Last updated: September 2026

1. Our AI position

   Escendant uses AI to help people draw out, organise, build and present memories. AI is assistive. It does not establish historical truth, prove that an event happened or turn an archive into a person. Check important output against original sources and lived knowledge before relying on or sharing it.

2. What AI does

   Feature	What it does	Main control
   Curator	Asks questions and suggests writing, structure, people, places, dates, themes and tags.	Edit, accept, reject, stop or delete.
   Transcription and analysis	Transcribes recordings, reads supplied files and extracts or suggests information.	Correct or delete; compare with the source.
   Search and context	Finds permitted Engrams and may search or summarise outside sources.	Open sources and reject or remove results.
   Generated and edited media	Creates or edits images, video, motion, emotes, avatars and speech.	Preview, regenerate, delete and keep the original.
   Escendant Chat	Answers questions about an archive subject using Engrams the questioner may access.	Permission-scoped sources, report and session controls.
   Safety support	May detect or flag content for security or moderation.	Report, block and complaint routes.

   Availability depends on age, role, consent, permissions, plan, credits, device and region.

3. How memory content is used

   Most AI processing happens through Escendant's backend. The Service sends the content and context needed for the requested task from Engrams and files the user is permitted to access. Some speech recognition, face detection, sensitive-media checks and quality checks may happen on the device.

   Indexing and retrieval used within Escendant is not general-purpose model training. We do not use private memories to train general-purpose AI models unless we first explain the separate use and obtain an explicit opt-in.

4. Providers and routing

   Escendant is not tied to one model. Different approved providers may handle language, transcription, web search, image, video, avatar or voice tasks. The backend may retry a failed task with another approved provider. This means more than one provider may process a request where needed for resilience, but not every provider receives every request.

   We approve routes for their purpose, data handling, locations, transfer safeguards, retention, training terms, security and deletion options. Our Privacy Policy explains provider categories and how to request current route information.

5. Errors and limitations

   AI may invent facts, merge people or events, infer a wrong date, location or relationship, mistranscribe speech, misread handwriting, omit context, reproduce bias, or create upsetting or anachronistic media. A confident tone is not proof.

   A completed Engram is the owner's account, not Escendant's independent verification. AI output is not legal, medical, financial, therapeutic, genealogical or historical advice and is not evidence of identity, consent, an event or a deceased person's wishes.

6. Escendant Chat

   Escendant Chat helps a permitted reader explore Engrams attached to one archive subject. It can connect information across those Engrams, cite them and open sources the reader is allowed to access. Answers may change with the question, permissions, available sources, retrieval results, model or safeguards.

   • Chat speaks about the subject in the third person, not as them.
   • It is not conscious and does not express the subject's present feelings or wishes.
   • It may be incomplete or wrong.
   • It should say when the permitted archive does not support an answer.
   • Inference and outside context should remain distinguishable from archive evidence. Absence from the archive is not proof that something did not happen.
   • A citation or source link is traceability, not independent verification, and remains subject to the reader's permissions.
   • Chat does not edit an Engram. A useful unanswered question may be offered to an authorised owner as a prompt for new Curator work.
   • Sharing an answer or transcript does not grant access to the underlying archive. Read-aloud uses a standard Curator voice, not a cloned subject voice.

7. Generated media, voice and likeness

   Generated media is a creative interpretation, not documentary proof. Image restoration can also alter details. Keep original material and do not present generated or materially altered output as an original record or evidence.

   Adult face and voice simulation is controlled by a separate opt-in and is not identity verification. It must not be used for a child. The user must have authority over the source and represented person's likeness or voice. See our Voice, Likeness and Biometric Data Policy.

   Escendant shows AI or synthetic-media context where the product provides a label or disclosure. Users must not remove it or misrepresent output. Some external platforms and exported formats may not preserve every label or item of metadata.

8. Human control and safety

   Depending on the feature, users can choose whether to start AI processing, control permitted sources, edit or reject output, correct transcripts, delete generated material, change sharing, disable optional simulation and report harmful output. Automated tools may assist moderation, but material restrictions can be challenged at complaints@escendant.ai.

   Under-18 accounts cannot create Weblinks, make Engrams Public or use voice and likeness simulation. We do not use private child memories for general-purpose model training.

9. Contact

   AI questions and error reports: support@escendant.ai
   Privacy and data rights: privacy@escendant.ai
   Security and child-safety concerns: security@escendant.ai
   Complaints and decision challenges: complaints@escendant.ai