How shubhamkunwartiwari/engramgraph works
EngramGraph is an interactive multi-agent graph-memory workspace and evaluation tool. Users work with agents that inspect a summary index, page relevant subtopics into working memory, update graph facts and relationships, and release context; the workspace also supports workflows, document-to-graph ingestion, saved paths, and benchmark runs. The server routes requests to the graph engine and agent orchestrator, while the React app presents graph and working-memory views. The README documents comparative retrieval benchmarks and OCC; some of those implementations are unsampled, so their internal relationships are not expanded here.
Main entry points: graphStore.ts Graph memory engine · App.tsx Workspace app · memory.ts Memory data model · agentOrchestrator.ts Agent orchestrator
Components 16
Workspace UI
- Workspace app
src/App.tsx - Graph canvas
src/components/HierarchicalGraphCanvas.tsx - Memory inspector
src/components/AgentWorkingMemoryInspector.tsx - Paging controls
src/components/AutonomousPagingOrchestrator.tsx - Memory dialogs
src/components/ModalsAndDrawers.tsx
API and agents
- HTTP API
server.ts - Agent orchestrator
server/agentOrchestrator.ts - Google Gemini
Graph memory
- Graph memory engine
server/graphStore.ts - Initial graph data
server/initialGraphData.ts - Browser fallback
src/utils/clientGraphFallback.ts - Memory data model
src/types/memory.ts
Evaluation
- Benchmark suite
server/graphStore.ts - Paper generator
src/utils/researchPaperGenerator.ts
Other components
- Workspace user
- FalkorDB
Connections 18
- Workspace user → Workspace app: uses
README.md - Workspace app → Graph canvas: renders
src/App.tsx - Workspace app → Memory dialogs: renders
src/App.tsx - Workspace app → Browser fallback: falls back to
src/App.tsx - Graph canvas → Memory inspector: shows
- Paging controls → Memory data model: uses types
src/components/AutonomousPagingOrchestrator.tsx - Workspace app → Memory data model: uses types
src/App.tsx - HTTP API → Graph memory engine: routes requests
server.ts - HTTP API → Agent orchestrator: runs agents
server.ts - Agent orchestrator → Graph memory engine: reads and updates
server/agentOrchestrator.ts - Agent orchestrator → Google Gemini: calls
server/agentOrchestrator.ts - Graph memory engine → Initial graph data: loads initial state
server/graphStore.ts - Browser fallback → Initial graph data: creates workspace
src/utils/clientGraphFallback.ts - Graph memory engine → FalkorDB: connects
- Graph memory engine → Benchmark suite: runs
server/graphStore.ts - Benchmark suite → Paper generator: supplies results
- Graph memory engine → Memory data model: uses model
server/graphStore.ts - Initial graph data → Memory data model: uses model
server/initialGraphData.ts