Built-in local AI (LLME)

Language-model assistance runs where you operate CryptPeer — mail, already-encrypted chat and notes — without pushing drafts through a public cloud AI API.

Clear promise

You get a language help layer (drafting, rephrasing, structuring, and sometimes short in-context summaries depending on usage) inside the relevant modules. Model inference runs in the CryptPeer footprint you operate (endpoint or infrastructure), not at a consumer SaaS AI vendor.

This is not a replacement for the web, legal research, or a search engine; it is not an autonomous agent executing actions on your behalf. It is practical writing productivity inside a controlled perimeter.

Where it plugs in: secure mail client (wording / polish), encrypted chat (support in the thread), collaborative notes (editing under deployment control).

Use cases

  • Sensitive mail: produce a first draft or soften tone before send, without copy-paste into an external chatbot.
  • Operational encrypted chat: phrase a reply inside the already end-to-end protected channel so context stays in one place.
  • Internal notes: reorganize bullets, list actions, or clarify rough notes in a space you govern.
  • Restricted environments: sites or desktops where public cloud AI is banned or filtered — keep minimal text assistance without opening network exceptions to consumer AI services.

Security reading

  • Cryptography unchanged: mail and chat keep their channel properties; local AI does not “add magic confidentiality” — it processes text inside the perimeter you chose.
  • Control: no systematic flow to a public AI service for this assistance; update policy, model choice, and hardware capacity remain your responsibility (default model sized for typical hardware; heavier models when resources allow).
  • Operational sovereignty: processing aligns with the same platform baseline as other modules (feature overview, messaging, transfer, sovereign translator for targeted multilingual work).

Bridging pages

Related sheets, product anchors, and decision-oriented pages.