A knowledge AI that's actually yours. Trained on your real data, nothing leaves your organization.
- 100%Processed on your own site
- 0%Data that leaves the organization
- 0THBToken cost per query (self-hosted, no cloud API calls)
If any of this sounds familiar
- Important knowledge is scattered across documents, photos, and the heads of people who've been there a long time
- You're not comfortable putting customer data or confidential contracts into ChatGPT or a cloud AI
- Per-token API costs spiral once staff start asking AI questions every day
- You want AI that actually understands your organization's own terms and context, not generic answers
How the system works
Import and read documents with on-device OCR
Documents, scans, and photos are converted to text by an OCR model that runs entirely on your own hardware — nothing is sent anywhere.
Retrieve and answer from real data with RAG
The system retrieves the relevant context from your existing knowledge before the chatbot answers. Answers are grounded in your organization's real documents, not generic internet knowledge.
Run whichever open model fits the job
We use an open-source language model suited to the task at hand, and can switch to a better one as they come out — you're not locked into one vendor or one model forever.
Fine-tune it for your organization once you're ready
Once there's enough real usage data, we fine-tune the open model with LoRA/QLoRA so it picks up your organization's own terms and context more accurately — without retraining a model from scratch.
The signature visual for this page has not been built yet (CT-031/032)
Technical specifications
- GPU used for testing
- NVIDIA RTX 6000 Ada
- Language model
- Open-source local LLM (swappable, not locked to one model)
- OCR
- Local OCR model (no external API calls)
Hardware used
Hardware package pricing is still being worked out — it depends on your data size and the model you choose. Get in touch for an estimate based on your actual needs.
See full specs and prices on the hardware pagePersonal data inside your organization's knowledge
If the data you import contains personal data about employees or customers, the system never sends data off-site, and supports access control over that data under PDPA.
Frequently asked questions
What this system cannot do
- Still in development, with no customer or pilot running it yet — there are no published accuracy or speed figures.
- Fine-tuning with LoRA/QLoRA needs a certain amount of real usage data before it shows a clear benefit — it isn't something that happens on day one.
- You need hardware on-site capable of running a large language model, which costs more up front than a cloud API — the trade-off is that no data leaves your organization and there's no per-call cost long-term.