Just Launched · GSD Graph by Open GSD

GSD GraphA modern, open source graph platform built for performance, flexibility, and scale.

Extract, normalize, connect, and query grounded local knowledge graphs from project docs, architecture decisions, and code — empowering AI coding agents with verified citations and zero cloud dependencies.

Run GSD Graph Instantly
Start MCP Server for Agents
gsd-graph · v0.1.0 · active
$ gsd-graph index ./docs ./src✓ indexed
› Documents: 64 files parsed
› Entities: 284 nodes extracted
› Relations: 612 directed edges resolved
› Storage: .gsd/graph.db (SQLite-vec)
MULTI-HOP GROUNDED ANSWER

"Worktree isolation in Milestone 07 enforces sandbox constraints from ADR-12 before test verification executes in harness.spec.ts."

Citations: ROADMAP.md:L42 · ADR-012.md:L18 · harness.spec.ts:L33
Architecture Core

Built for Performance, Flexibility, & Scale

Knowledge graphs that don't require heavy cloud infrastructure, subscription fees, or complex cluster setups.

High Performance

Sub-millisecond local vector and graph traversals. Built to respond instantly inside agent execution loops without latency spikes.

Verified

Developer Friendly

Zero-config setup. Drop into any repository with 1 line, inspect with intuitive query syntax, or connect directly via MCP.

Verified

Scalable by Design

Local-first SQLite + vector indexing that scales effortlessly from small side projects to multi-million node monorepos.

Verified

100% Open Source

MIT licensed, completely local, zero telemetry, and zero cloud vendor lock-in. Your codebase knowledge remains strictly yours.

Verified
Live Visualizer

Explore GSD Graph in Action

Click nodes or execute pre-configured queries to see how GSD Graph maps and traverses complex codebase knowledge.

GSD GRAPH RUNTIME

Interactive Grounded Knowledge Graph

Local Vector-Graph DB Connected
MILMilestone 07SPEAuth & ScopesDECADR-12 WorktreesCODengine/runtime.tsCHETest Suite & Proof
Click any node to inspect grounded citations5 nodes · 6 relations
CYPHER / VECTOR QUERY0.42ms execution
MATCH (m:Milestone)-[r*1..3]->(c:Check) RETURN m, r, c
Found 2 paths linking Milestone 07 to Test Suite & Proof via runtime.ts
milestoneID: milestone-07

Milestone 07

Safer auto mode execution and worktree isolation for agentic coding.

Source File
ROADMAP.md
Line Span
L42-L58
Connected Graph Edges
milestone-07:DEFINES_REQUIREMENTSspec-auth
milestone-07:GUIDED_BYadr-worktree
milestone-07:VALIDATES_GOALverify-suite
Zero cloud telemetry · Pure local SQLite-vec engine
The Grounding Engine

How GSD Graph Works

A deterministic 5-stage pipeline turning unstructured files into a rich, queryable knowledge network for developers and AI agents.

01

Ingest & AST Parse

Scans markdown specs, ADRs, codebases (TypeScript, Python, Rust, Go), and git commit histories without hallucinations.

02

Entity & Relation Extraction

Identifies components, APIs, milestones, decisions, invariants, and dependencies with exact source file line-spans.

03

Graph Normalization

Resolves aliases and cross-document references into a unified topological graph (:DEPENDS_ON, :IMPLEMENTS, :VERIFIES).

04

Local Vector Grounding

Pairs relational graph structure with local fast embeddings for hybrid topological and semantic searches.

05

Agent Context Synthesis

Feeds precise subgraphs and verified line-level citations into Claude Code, Cursor, Codex, and gsd-pi prompts.

Seamless Integration

Native MCP Server for AI Coding Agents

Plug GSD Graph directly into Claude Code, Cursor, Windsurf, Copilot, or gsd-pi. Agents can inspect entities, run path traversals, and cite line-level evidence without context blow-up.

Zero-overhead MCP Tools: Exposes graph_query, find_related, and get_citations.
Line-numbered Grounding: Every node maintains exact file paths and source span offsets.
Local & Secure: Runs entirely on your machine with no data leaving your environment.
claude_desktop_config.json / mcp.jsonMCP v1
{
  "mcpServers": {
    "gsd-graph": {
      "command": "npx",
      "args": ["-y", "@opengsd/gsd-graph@latest", "mcp"],
      "env": {
        "GSD_GRAPH_ROOT": "."
      }
    }
  }
}
Comparison

Why GSD Graph?

How GSD Graph compares to traditional hosted graph databases and raw vector search.

FeatureGSD Graph (Open GSD)Hosted Graph DBs (Neo4j, etc.)Raw Vector Search (Pinecone, etc.)
Local / Offline First✓ 100% Local (SQLite-vec)Requires Server / CloudCloud Hosted API
Setup Complexity✓ 1-line (`npx @opengsd/gsd-graph`)Docker, JVM, Cluster ConfigAPI Keys, Billing, Index Setup
Multi-Hop Traversal✓ Native Topological QuerySupported via Cypher❌ No relation awareness
Line-Level Citations✓ Exact File & Line SpansManual schema mappingCoarse document chunks
AI Agent MCP Integration✓ Built-in stdio / SSE MCP ServerCustom adapters neededCustom wrapper needed
License & Cost✓ Free & MIT LicensedEnterprise tiers / high costPer-query / storage fees
Get Started Today

Build Your First Knowledge Graph in Seconds

Join open-source developers who are making agentic coding grounded, deterministic, and verifiable.