Aethos · RAG · by STK Engineering Sovereign knowledge retrieval · 2026

Your knowledge.
Instantly accessible.

RAG across your entire document landscape — Confluence, SharePoint, wikis, tickets, runbooks. Answers in natural language, with source citations, on your hardware.

I · CAPABILITIES Not search. Understanding.

Not search.
Understanding.

A

Multi-Source Ingestion

Connect Confluence, SharePoint, Git repos, file shares, databases, ticket systems. One unified knowledge layer across all silos. Incremental sync, not batch.

B

Semantic Search

Vector embeddings with hybrid BM25 ranking. Understands intent, not just keywords. "How do we handle GDPR deletion requests?" returns the actual process, not 200 documents.

C

Source-Cited Answers

Every response includes clickable source references. Know where the answer came from. Audit trail for compliance. No hallucination without citation.

D

Access Control Inheritance

Respects your existing permissions. SharePoint ACLs, Confluence spaces, LDAP groups. A user only sees answers from documents they are allowed to read.

E

Domain Adaptation

Fine-tune retrieval for your terminology, your abbreviations, your internal naming conventions. Banking, healthcare, legal, engineering — each domain has its own language.

F

On-Premise Embeddings

Local embedding models (BGE, E5, Nomic). No document content leaves your network. Vector store on your infrastructure.

II · USE CASES Where knowledge creates value

Where knowledge
creates value.

CASE 01

Internal Knowledge Portal

Self-service answers for employees across HR policies, IT runbooks, product documentation, engineering decisions. Reduces tier-1 support tickets by 60%+. Makes new hires productive from week one.

CASE 02

Regulatory Compliance Assistant

Instant answers across compliance frameworks, internal policies, audit documentation. "Does our data retention policy comply with NIS-2?" — answered in seconds with source links.

CASE 03

Customer Support Augmentation

Support agents get instant answers from product docs, known issues, resolution histories. Faster resolution, consistent quality, automatic documentation of each case.

CASE 04

Engineering Knowledge Base

Architecture decisions, API documentation, deployment runbooks, incident post-mortems. The institutional memory that doesn't leave when engineers do.

III · THE PIPELINE From question to cited answer

From question to
cited answer.

Every query follows a deterministic six-stage pipeline. Each stage is observable, each result is traceable, every source is verifiable.

I

Query Input

User submits a natural-language question through the interface or API.

II

Intent Analysis

Parses query intent, extracts entities, determines scope and required source types.

III

Hybrid Retrieval

Parallel vector search and BM25 keyword retrieval across all connected sources.

IV

Re-Ranking

Cross-encoder re-ranking of candidate chunks. Filters by ACL permissions and relevance.

V

LLM Synthesis

Local LLM generates a coherent answer grounded in retrieved context. No fabrication.

VI

Cited Response

Answer delivered with inline source citations, confidence scores and audit metadata.

IV · SPECIFICATIONS Technical depth

Built for
production.

Retrieval

Hybrid vector + BM25

Sub-500ms query time. Parallel dense and sparse retrieval with cross-encoder re-ranking for maximum precision.

Embeddings

BGE / E5 / Nomic

Local inference, 1024-dimensional vectors. No external API calls. ONNX Runtime acceleration on CPU and GPU.

LLM

Llama 4 / Qwen 3.6 / Gemma 4 / Nemotron

Local inference, fully configurable. Choose the model that fits your hardware, your language, your domain.

Sources

Confluence, SharePoint, Git, Databases

File shares, SQL/NoSQL databases, REST APIs. Incremental sync with change detection. Plug in any source via connector SDK.

Security

Zero-cloud, ACL inheritance

Full audit trail. Permission-aware retrieval. No data leaves your network. LDAP/AD integration out of the box.

Scale

10M+ documents

Incremental sync, horizontal scaling. Shard vector stores across nodes. Production-tested at enterprise scale.

Sovereign by design

Your data stays yours.

No document content ever leaves your network. Aethos RAG runs entirely on your infrastructure — embeddings, vector stores, LLM inference. Enterprise-grade knowledge retrieval, fully sovereign.

V · GET STARTED From question to deployment

Begin where knowledge
matters most.

Ready to make your organisation's knowledge instantly accessible? Tell us about your data landscape and we'll design a pilot tailored to your sources.

Vienna, Austria
office@stk-engineering.com
Ferrogasse 59, 1180 Wien
Belgrade, Serbia
office@stk-engineering.com
Moravska 6, 11000 Beograd
Chalandri, Greece
office@stk-engineering.com
Nestoros 1, 15231 Chalandri
I am interested in