Evidence-first retrieval for engineering truth
Iconys Knowledge Engine
Open-Source Enterprise RAG. CAD-first by design.
Open-source under Apache 2.0
Iconys Knowledge Engine is an evidence-first retrieval platform for engineering products that need grounded answers, citations, and operational control. It combines hybrid retrieval, graph-aware reasoning, structured SQL intents, and project memory in one Rust runtime.
Designed first for CAD, standards, materials, tolerances, and long-lived project knowledge, it is equally usable for any high-trust domain where answers must be explainable, tenant-aware, and deployable on your own infrastructure.
The source repository is public on GitHub, and the official container image is published on Docker Hub for teams that want a faster self-hosted start.
Repository
github.com/iconys-bg/iconys-knowledge-engine10-Minute Quickstart
github.com/iconys-bg/iconys-knowledge-engine#10-minute-self-hosted-quickstartdocker compose up -d
Deployment Model
Self-hosted, containerized, and cloud-portable for regulated teams that cannot outsource their knowledge layer.
Operational Goal
Answers with evidence, memory, and policy boundaries instead of generic chat completions.
Hybrid Retrieval
Dense + lexical retrieval with RRF fusion over pgvector and Tantivy instead of relying on a single retrieval path.
Graph-Aware Reasoning
Apache AGE traversal for relationship-heavy questions such as dependencies, references, supersession, and connected facts.
Structured Intents
Deterministic SQL paths for standards, materials, tolerances, and other high-value engineering lookups.
Evidence Packs
Answers are assembled with citations, confidence, conflict tracking, and verification hooks instead of opaque chat text.
Memory + Tenancy
Project memory, tenant isolation, RBAC-aware retrieval, and audit trails keep product context durable and segmented.
Feedback + Healing
Continuous evaluation, knowledge gap tracking, and self-healing loops turn misses into an observable improvement system.
CAD-first, Not CAD-only
Built first for engineering truth: standards, materials, tolerances, references, and long-lived project memory. Reusable for any high-trust knowledge workload.
Beyond Basic RAG
Vector search alone is not enough. Knowledge Engine combines hybrid retrieval, graph traversal, structured lookups, and evidence assembly in one system.
Open Source, No Lock-in
Apache 2.0 licensing means your knowledge layer stays portable, inspectable, and controllable by your own team.
Product-Ready Interfaces
REST and NATS surfaces let product teams plug retrieval, evidence, and orchestration into real applications instead of demo chat shells.
Multi-Tenant by Design
Tenant IDs flow through retrieval, memory, storage, and policy enforcement so enterprise isolation is part of the runtime, not an afterthought.
Operational Maturity
Docker, GHCR, observability hooks, and a Rust runtime make the system deployable in production environments that care about latency and control.
Ready to build on a serious RAG foundation?
Open-source, CAD-first, multi-tenant, and evidence-first.