Point Aryx at your data. Get a deduplicated, linked knowledge graph you can ask questions of — with provenance on every answer.
Install · Docker Hub · User guide · Quickstart data · Contributing · Raven (Claude Code) · Code of conduct · Licensing
Public · source-available (BSL 1.1). Postgres is the system of record; the live knowledge graph is projected into FalkorDB.
Aryx turns the data you already have into a workspace-scoped knowledge graph you can ask questions of. You state goals in plain English, connect a database or upload files, approve the model Aryx proposes, resolve duplicates into golden entities, and explore with Ask, Model, Data (tree / table / entity graph), and Accuracy Lab — with provenance on answers and merges.
| Edition in this repo | Aryx Lite — single-team outcome mapping (laptop / small server) |
| Not (yet) | Multi-tenant governed enterprise estate (that path is Enterprise / Aryx-o) |
| SoT / graph | PostgreSQL truth · FalkorDB rebuildable projection |
| UI | Next.js only (apps/web) — no Streamlit |
| Images (public) | giggsodocker/aryx-lite (API · worker · MCP) · giggsodocker/aryx-lite-web (Next.js UI) |
| Tags | latest · 1.0.0 · v1.0.0 · git short SHA (same tags on both images) |
| License | BSL 1.1 → GPL-3.0-or-later on 2029-07-15 · details |
| Docker Hub overview | docs/DOCKERHUB.md · aryx-lite · aryx-lite-web |
| Step | What happens |
|---|---|
| 1. Goals | You describe what you want to figure out in plain English |
| 2. Ingest | Connect a database or upload files (CSV, PDF, DOCX, …) |
| 3. Resolve | Duplicates merge into golden entities; weak pairs wait for human review |
| 4. Link | Cross-file relationships are discovered and projected into a graph |
| 5. Explore | Ask, Model canvas, Data explorer (tree / table / entity graph), Accuracy Lab |
Built for a single team’s outcome mapping on a laptop or small server — not yet a multi-tenant governed enterprise estate.
Requirements: Docker + Docker Compose · ~8 GB RAM recommended (Ollama models)
Images (Docker Hub, public — no login to pull):
| Image | Role | Tags |
|---|---|---|
giggsodocker/aryx-lite |
API · worker · MCP | latest · 1.0.0 · v1.0.0 · <sha> |
giggsodocker/aryx-lite-web |
Next.js UI | same tags |
git clone https://github.com/giggsoinc/aryx.git
cd aryx
cp .env.example .env # edit passwords / LLM keys as needed
docker compose pull # pulls public Hub images (no docker login)
docker compose up -d # builds from source only if pull/build policy requires itFirst boot pulls LLM models into Ollama (can take several minutes).
| Surface | URL |
|---|---|
| Web UI | http://localhost:3000 |
| Settings (LLM provider / API keys) | http://localhost:3000/settings |
| API docs | http://localhost:8088/docs |
| MCP (SSE) | http://localhost:8765/sse |
Then: open the web UI → New workspace → Onboard (/start) → goals → add a source → run.
Sample data: upload the CSVs in examples/quickstart/ (customers + tickets) to exercise multi-file ingest and the entity graph.
Full steps, ports, and troubleshooting: docs/INSTALL.md · Hub copy: docs/DOCKERHUB.md · UI walkthrough: docs/USER_GUIDE.md
curl -s http://localhost:8088/health
# then open http://localhost:3000docker pull giggsodocker/aryx-lite:1.0.0
docker pull giggsodocker/aryx-lite-web:1.0.0
# or latest / v1.0.0 / git short SHA| Route | Purpose |
|---|---|
/start |
Guided onboard wizard |
/ |
Ask — grounded Q&A with citations |
/model |
Ontology canvas (types, relationships, survivorship, axioms) |
/data |
Transparency explorer — Tree, Table, interactive entity Graph |
/lab |
Accuracy Lab — ontology ON vs OFF + reasoner check |
/settings |
LLM provider — Ollama, Anthropic, OpenAI-compatible, Gemini, Grok (xAI) |
There is no Streamlit UI. The product UI is Next.js only.
- Multi-source ingest — Postgres, MySQL, Oracle; files (CSV/JSON/PDF/DOCX/PPTX/images)
- Discovery-driven ontology — propose types, human approval gate, RDF/OWL import-export
- Entity resolution — multi-key blocking, four-band scoring, HITL review, survivorship policies
- Cross-file relationships — deterministic FK discovery after multi-file upload
- Entity graph — pan/zoom, search, type filters, click-to-explore neighbors + detail panel
- Workspace isolation — LIST-partitioned Postgres; one graph per workspace
- Local or cloud LLMs — default Ollama; swap live under Settings (no restart)
- MCP — tools over SSE for external agents
- Ports & adapters — relational / graph / vector / LLM / reasoner / compute swappable for Enterprise / Aryx-o
| Layer | Technology |
|---|---|
| API | FastAPI + OpenAPI |
| Web | Next.js 15 (App Router, Tailwind) |
| Database (source of truth) | PostgreSQL 16 + pgvector |
| Graph projection | FalkorDB (one named graph per workspace) |
| LLM | Ollama (default) · Anthropic · OpenAI-compatible · Gemini · Grok |
| Agents | MCP over SSE |
| Deploy | Docker Compose · giggsodocker/aryx-lite |
Aryx is an application on top of FalkorDB (and Postgres), not a fork of the database.
| Doc | Contents |
|---|---|
| Install | Docker, Hub images, env, security, updates, local dev |
| User guide | Workspaces, onboard, Ask, Data, Model, Lab, Settings |
| Quickstart data | Sample CSVs for multi-file upload → graph |
| Features | Capability matrix |
| Architecture | System design |
| Licensing | BSL plain English |
| Ingestion | Deep ingest walkthrough |
| Docker Hub overview | Image description (API / worker / MCP) |
| Raven | Optional Claude Code workflow (.claude + manifest) |
| Benchmarks | ER measurements |
Diagrams: Business view · Technical flow
- Code of Conduct
- Contributing
- Security policy
- Issues · Discussions (if enabled)
Source-available under the Business Source License 1.1. You may use and modify Aryx Lite for evaluation, research, and internal production under the Additional Use Grant. Competing multi-tenant hosting of this work requires a commercial license until the Change Date (2029-07-15), when this version becomes GPL-3.0-or-later.
Full text: LICENSE · Summary: docs/LICENSING.md · Notice: NOTICE
Maintained by Giggso.