Tiranyx

/

sidix-lora

like 0

Hey everyone,

I built SIDIX — a free, open-source AI agent that runs 100% locally on your own server. No API fees, no data leaving your machine, no monthly subscription.

But I want to be honest about what makes it genuinely different — not just marketing.

What SIDIX does that closed models don''t:

1. You can audit every inference decision — the ReAct loop is fully visible, every tool call is logged
2. Epistemic labeling on every output — [FACT], [OPINION], [SPECULATION], [UNKNOWN]. Hallucinations are labeled, not hidden.
3. Citation chain (Sanad) — every factual claim traces back to a source chunk. You know why it said something.
4. It learns from your usage — accepted outputs feed a weekly prompt optimizer. The more you use it, the better it gets for your domain.
5. True self-hosting — if Ollama crashes, SIDIX returns an honest error. It does NOT silently fall back to GPT/Gemini.

What it can''t do (yet, honest):

• Voice / video (Sprint 7)
• Real-time web browsing at GPT-4 speed
• The base is Qwen2.5-7B — smaller than GPT-4o, obviously

Questions I''d love community input on:

• What use cases would you most want a self-hosted agent for?
• Anyone tried running it on < 8GB RAM? What was the experience?
• Any Indonesian NLP researchers here? SIDIX has a strong Indonesian corpus — would love collaborators.

Full repo: https://github.com/tiranyx/sidix
Live demo (free): https://app.sidixlab.com

Happy to answer anything.