Dokumentasi/Ecosystem
AI on the server
A language model that runs without third-party services.
The language model in this ecosystem runs on our own server. No prompt or user data is sent to an external AI provider.
How it is put together
| Part | Detail |
|---|---|
| Interface | Tark AI — tark.aflos.xyz |
| Model | Qwen3 1.7B |
| Inference engine | Ollama, listening only on 127.0.0.1 |
| Web search | SearXNG, self-hosted |
| Data path | Never leaves the server |
Why it runs locally
- Privacy. Your questions are not sent to a third party, not logged by another service, and not used to train anyone's model.
- No external dependency. The assistant keeps working even if a connection to an outside AI provider is having problems.
- Fixed cost. There is no per-request charge that varies with usage.
Self-hosted search
Web search uses SearXNG, installed on the same server. Search queries are not forwarded to a third-party search engine, so no search profile is built up.
⚠️
A model with 1.7 billion parameters performs
differently from commercial models many times larger. Its advantage is privacy and cost,
not breadth of knowledge or depth of reasoning.
Limitations
- The model can be confidently wrong. Answers that matter should be checked.
- Its knowledge coverage is limited and not always current.
- For questions needing up-to-date data, the assistant has to search the web first.
Apakah halaman ini membantu?