Open Source · GraphRAG · MIT License

If you can't trace it,
you can't trust it.

TraceGraph is a GraphRAG citation explorer that grounds every AI answer in a verifiable knowledge graph. See the entities, follow the relationships, trace the truth.

Built with

L
LightRAG
F
FastAPI
N
Next.js
R
React
O
OpenAI
X
NetworkX
TS
TypeScript
T
Tailwind
D
Docker
P
Pydantic
L
LightRAG
F
FastAPI
N
Next.js
R
React
O
OpenAI
X
NetworkX
TS
TypeScript
T
Tailwind
D
Docker
P
Pydantic
Traditional RAG

Vector search finds text that looks like the answer.

Standard RAG retrieves document chunks based on semantic similarity. It works for simple lookups but fails when questions require connecting information across multiple documents, understanding entity relationships, or providing auditable citation trails.

Query
Vector DB
Chunks
LLM
Answer without source trail
GraphRAG

Graph traversal finds information that is related to the answer.

GraphRAG builds a knowledge graph from your documents — entities, relationships, and community structures. Every answer traces back through the graph to its source, providing the citation chain that regulated industries demand.

Query
Graph + Vectors
LLM
Answer + Citations + Entity Chain
0
Entities Extracted
0
Relationships Discovered
0
Documents Ingested
0
Search Modes Available
Features

Everything you need to trace AI reasoning.

From entity extraction to citation trails, TraceGraph provides the complete toolkit for explainable AI.

CORE FEATURE

Full Citation Trail

Every AI answer traces back through entity chains to source documents. Click any citation to see the exact graph path that led to each claim.

clinical_guidelines.pdf
GraphRAGKnowledge GraphStructured Grounding

"GraphRAG reduces hallucinations by grounding responses..."

FLEXIBLE

4 Search Modes

Hybrid, Local, Global, or Naive — choose how to traverse the knowledge graph for each query.

Hybrid
Local
Global
Naive
COMPARISON

Side-by-Side

Compare naive vector search against graph-enhanced retrieval on the same question.

Naive RAG
No graph
GraphRAG
+ Citations
VISUAL

6 Entity Types

Concepts, technologies, organizations, regulations, persons, and documents — each color-coded in the graph.

Concept
Tech
Org
Reg
Person
Doc
INSTANT

Works Offline

Ships with sample graph data. No API keys needed to explore the UI.

Demo Mode Active
COMFORTABLE

Resizable Panels

Drag to resize the graph, AI response, and citation trail panels. Read long answers without feeling cramped.

Graph
Response
Citations
How it Works

From documents to traceable answers in three steps.

01

Ingest Documents

Upload PDFs, text files, or any document corpus. TraceGraph chunks and embeds them for retrieval.

clinical_guidelines.pdf
research_paper.txt
regulations.md
02

Extract Knowledge Graph

LLMs automatically extract entities and relationships, building a structured knowledge graph with community hierarchies.

03

Query with Citations

Ask questions in natural language. Get answers grounded in the knowledge graph with full citation trails.

How does GraphRAG reduce hallucinations?

GraphRAG grounds responses in verified entity relationships...

[1][2][3]
Developer Experience

Query the knowledge graph in three lines.

import httpx

response = httpx.post("https:">//tracegraph-ls2t.onrender.com/query", json={
    "query": "How does GraphRAG reduce hallucinations?",
    "mode": "hybrid"
})

result = response.json()
print(result["answer"])                    "text-[#6a737d]"># Grounded AI response
print(result["citations"][0]["entity_chain"])  "text-[#6a737d]"># ['GraphRAG', 'Knowledge Graph', 'Structured Grounding']
print(f"Sources: {len(result['citations'])}")  "text-[#6a737d]"># Sources: 10
Architecture

Designed for clarity, built for scale.

Browser
Frontend
Next.js 16App Router
React 19Server Components
Force GraphVisualization
Backend
FastAPIREST API
LightRAG 1.4Graph Engine
OpenAI APILLM Provider
Storage
GraphMLVector DBKV Stores
Py
Python
FA
FastAPI
LR
LightRAG
OA
OpenAI
Nx
Next.js
Re
React
TS
TypeScript
Dk
Docker
Use Cases

Built for domains where accuracy is non-negotiable.

Healthcare Decision Support

Knowledge graphs over clinical guidelines, drug interactions, and patient data. UMLS and SNOMED CT entity support.

3 corpus documents

Pharmacovigilance

VAERS adverse event analysis with cross-document reasoning. Brighton Collaboration case definition support.

2 corpus documents

EU AI Act Compliance

Article 13 transparency and Article 14 human oversight requirements met through built-in citation traceability.

2 corpus documents

Clinical Trial Analysis

Multi-hop reasoning across trial data, drug efficacy studies, and safety profiles. ClinicalTrials.gov integration.

2 corpus documents

Trace the truth.

Open source. MIT licensed. Ready to deploy.