GS
GraphSentinel
Launch Console
FalkorDB / Context Engine for Autonomous Agents

Vector databases give agents memory.
Graphs give them an architecture map.

When production microservices fail, the root cause is rarely the service throwing the 500 error. GraphSentinel models infrastructure topology, team ownership, and past runbooks in FalkorDB—enabling autonomous agents to traverse multi-hop blast radii and resolve outages without guessing.

Launch Interactive Console → Test Live Incident Simulator
Traversal Latency 0.94 ms P99 In-memory pointer hops
Query Language OpenCypher Deterministic path match
Cross-Session State Episodic Graph Persistent incident history
Action Interface MCP Tools Automated runbook dispatch
The Problem with Vector RAG

Why vector similarity is the wrong tool for system infrastructure

Vector databases calculate text cosine distance. But infrastructure dependencies aren't semantic similarity problems—they are directed acyclic graphs.

What Vector RAG Sees During an Outage Cosine Match

When payment-service throws 504 timeouts, a vector search retrieves fragmented wiki chunks mentioning "payments", "Stripe API", or "credit card".

• Similarity: 0.81 → "Billing Policy v2.pdf"
• Similarity: 0.77 → "How to configure Stripe webhooks"
× Result: The agent has no idea that order-service depends on payment-service, or that the failure will take down the API gateway.
What GraphSentinel Sees in FalkorDB Deterministic Pointer

The agent runs a single Cypher traversal following physical edges across the microservice mesh in 0.94 milliseconds.

MATCH (target:Service {id: 'payment-service'})
<-[:CALLS*1..3]-(upstream:Service)
RETURN upstream.id, length(path)
✓ Exact blast radius: [order-service → api-gateway].
✓ Recalled Incident #INC-2026-0921 mitigation runbook.
Live Simulator

Test the blast radius engine right here

Select an incident to simulate the agent's graph traversal:

Open Fullscreen Console →
Trigger Incident:
falkordb: ready
1. Executed FalkorDB Cypher Query
MATCH path = (dependent)-[:CALLS|ROUTES_TO*1..3]->(target {id: 'payment-service'})
RETURN dependent.id AS ImpactedNode, length(path) AS Depth;
Execution Metrics Engine: FalkorDB (In-Memory Graph Engine) Traversal Latency: 0.94 ms
2. Resolved Topology & MCP Action
Cascading Blast Radius (Upstream Impact): • order-service (1 hop • Tier-1)
• api-gateway (2 hops • Tier-0 Ingress)
Episodic Memory Recall (Track 02): Recalled Incident #INC-2026-0921: "Tripped circuit-breaker on order-service; rerouted to cache."
Dispatched MCP Action: Tool: k8s_isolate_pod(target='payment-service') • Status: Executed
Developer Experience

Simple Python SDK integration

GraphSentinel interacts with FalkorDB using the standard Python client. No proprietary query wrappers or black-box abstractions.

# Install: pip install falkordb from falkordb import FalkorDB # Connect to FalkorDB instance client = FalkorDB(host="localhost", port=6379) graph = client.select_graph("production-mesh") # Run multi-hop blast radius traversal inside the agent loop query = """ MATCH path = (root:Service {id: $service})<-[:CALLS*1..3]-(upstream:Service) RETURN upstream.id AS affected_service, length(path) AS hop_depth """ result = graph.query(query, {"service": "payment-service"}) for record in result.result_set: print(f"Impacted: {record[0]} at hop {record[1]}")
Graph Hacks 2026

Architected for all three hackathon tracks

Track 01 Prize: iPhone 18 Pro

Agents That Act on Connected Data

Autonomous multi-hop reasoning. The agent executes Cypher path algorithms to calculate cascading blast radii and triggers MCP tools based on topology.

✓ Verified multi-hop reachability
Track 02 Prize: PS5 + GTA VI

Agent Memory & Coordination

Persistent cross-session episodic memory. The agent stores incident resolutions as new graph edges, reusing verified runbooks with multi-tenant isolation.

✓ Cross-session episodic graph memory
Track 03 Prize: M6 Mac Mini

Company Brain

Connects microservices, CVEs, team ownership, and on-call rotations (Team → Lead → Pager) with exact citations and provenance.

✓ Organizational ownership graph
Turnkey Deployment

Start FalkorDB & GraphSentinel in 60 seconds

# Clone repo & launch via Docker Compose
git clone https://github.com/example/graph-sentinel.git
cd graph-sentinel && docker compose up -d

# Active services:
• GraphSentinel UI & API: http://localhost:8080
• Investigation Console: http://localhost:8080/app.html
• FalkorDB Browser: http://localhost:3000
• FalkorDB Cypher Port: localhost:6379
Open Investigation Console →