Key points at a glance
- Don't just look at "how human it sounds" — pay more attention to memory, proactivity, data, and model.
- Memory should be long-term and viewable / editable / deletable; proactivity should be adjustable and non-intrusive.
- Ideally data can be self-hosted and controlled, and the model can be swapped.
First look at memory: is it truly long-term, and is it controllable
Ask yourself: can it remember specific things you've said across sessions and across months? Or does it forget the moment a session ends? Truly useful companionship depends on long-term memory. Then look at whether you can view, edit, and delete that memory — controllable memory is what makes it trustworthy, rather than an opaque black box.
Next look at proactivity: is it intrusive, and can it be adjusted
Good companionship reaches out at the right moments instead of passively waiting for you to speak — but it must never become harassment. Check whether its proactivity is measured, and whether you can mute it when you'd rather be left alone. Proactive without overstepping is the dividing line between a good experience and a bad one.
Finally look at data and model: can it be self-hosted, and can it be swapped
Private conversations are best kept in your own hands: prefer solutions that are open source, self-hostable, and let you export and delete your data. Then see whether the model can be swapped — supporting both cloud and local means you can freely balance quality against privacy. On all four points, Enclave hands the choice back to you.
FAQ
Does Enclave's AI really have long-term memory?
Yes. Enclave uses a structured long-term memory system rather than a simple context window, so experts can still cite specific things you said months ago — not a vague "I think you mentioned that." Every memory is yours to view, edit, or delete at any time.Will the AI experts proactively reach out to me?
Yes. Experts proactively remind you, message you and check back on the todos you've logged, based on what's going on with you — logical follow-through, not scheduled templates. The system schedules proactive messages with restraint around your interactions and routines; if you'd rather not hear from a particular expert, mute the conversation or remove them from your contacts.Is my conversation data safe? Will it be used for training?
In self-hosted mode, your conversations live only on your own drive — they never leave your machine and are never used by any third party for training. The model layer can connect to cloud APIs like OpenAI and Anthropic, or be swapped for local Ollama / vLLM to run fully offline; you decide which kind of conversation goes to which model.
Ready to try it?
Open it in your browser — no credit card, no install.