The cost of handing your private conversations to the cloud
Mainstream AI chat products store your conversations in the cloud and feed them into future training. Even when the ToS says "we won't," you have no way to verify it yourself. The moment your account gets locked, the platform shuts down, or compliance rules shift, those chats are no longer in your hands.
The entire stack is open-source — running it yourself is the strongest privacy promise there is
The entire Enclave monorepo lives on GitHub under MIT — audit it, modify it, fork it. A single docker compose spins up your own instance with API, frontend, database, and vector index all running locally. The model layer mixes cloud APIs and local Ollama / vLLM however you want — you decide which model handles which kind of conversation.
Side-by-side comparison
| Dimension | Enclave | SillyTavern |
|---|---|---|
| Product positioning | An out-of-the-box private assistant world | A highly customizable roleplay chat frontend; bring your own backend |
| Open source & self-hosting | MIT open source, self-hostable | Open source, self-hostable (self-hosted frontend) |
| Long-term memory | Built-in structured long-term memory, ready out of the box | Configure it yourself via extensions / vector plugins |
| Proactivity | Experts proactively remind and message you — measured, never spammy | Primarily passive conversation |
| Multiple characters & relationships | Multiple experts and a relationship network, with group chats orchestrated by the system | Supports multiple characters, but requires manual orchestration |
| Real deliverables | Can directly produce PPT / Word / Excel | Focused on roleplay and conversation |
How it works in practice
Fully open-source code
MIT-licensed with no binary black boxes — every "we won't touch your data" promise is something you can audit yourself.
Three-minute Docker deployment
clone → cp .env → docker compose up. The README spells out the whole flow up top — anyone can spin it up.
Fully swappable models
Plug in OpenAI, Anthropic, Google, DeepSeek, or local Ollama / vLLM — different characters can even run on different models.
Data you can back up and delete
Self-hosted, the data itself is files on your disk: copying is backing up, taking them along is migrating, deleting is truly deleting — no "copies you can't remove".
FAQ
How hard is self-hosting?
If you've used docker compose, it's about as hard as running any other web service. The README's "three-minute deploy" flow is real — clone, edit .env, start the services. Three steps.Can it run fully offline?
Yes. Swap the model layer for local Ollama or vLLM, turn off real-world sync, and the entire system stops making any outbound request.Can a self-hosted instance still get the official feature updates?
Run git pull + docker compose up -d to roll forward to the latest version. The CHANGELOG flags breaking changes for every release.Can I self-host Enclave? Is it hard?
Yes, and it's about as hard as running an ordinary web service. If you've used docker compose, three steps get you your own instance: clone the repo, copy .env, and docker compose up. The API, frontend, database, and vector index all run on your own machine.Which large language models does Enclave support?
Enclave's model layer is fully swappable: OpenAI, Anthropic, Google, DeepSeek, and local Ollama / vLLM can all be configured. Different characters can even use different models, freely allocated by scenario and cost.What do I need to self-host Enclave? Do I need a GPU?
Not necessarily a GPU. If you use cloud model APIs (OpenAI, Anthropic, etc.), an ordinary server or home machine that can run Docker is enough — inference happens in the cloud, and your machine only runs the app and database. You only need the corresponding VRAM and compute if you want offline inference with local models (Ollama / vLLM).
Ready to try it?
Open it in your browser — no credit card, no install.