Tacita documentation
A curated subset of Tacita's engineering wiki. Each page is the citable, in-detail account of one subsystem and is updated when the implementation moves.
Documentation here is intentionally narrow. Tacita's full engineering wiki is much larger and tracks internal work; the pages below are the public, stable subset we are happy to be cited on. If you find a discrepancy between a doc page and the app's behaviour, the app wins and the page will be corrected.
- Privacy architecture How the encryption stack actually works. Argon2id (m=64 MiB, t=3, p=4) for new vaults with PBKDF2-HMAC-SHA256 on legacy vaults, AES-256-GCM at rest behind a versioned envelope header, the master key in RAM only, secureDelete, and the per-vault encrypted blob store.
- Character cards Native CCv2, CCv3, and CHARX import. The full field map, the lorebook activation algorithm, sanitisation, caps, macros, and round-trip preservation.
- Models The two curated Gemma 4 GGUFs, Light (E2B) and Pro (E4B). Device tier matrix, RAM requirements, KV-cache quantisation policy, and how the runtime planner picks settings.
- Device requirements The hardware floor Tacita ships against. Flagship-tier Android with Vulkan 1.2 and 8 GB of RAM, or iPhone 15 Pro and newer on iOS 17. Where the bar is, why it is there, and what happens if your phone is below it.
Direct answers to product questions: FAQ. Definitions of the technical terms used here: glossary. How Tacita compares to a typical cloud chatbot app: compare.