Registration
Registration is free and takes less than a minute. Enter on your own or with a team of up to four, and stay registered right up until submissions close.
Pick a question that lives in the relationships, and build the product that answers it.
Enter on your own or with up to three others. Register once, either way.
Free to enter, online from anywhere, and no graph database experience assumed.
Investigation and risk, agentic AI, or the GraphRAG SDK. Enter the one your idea fits.
01 / The challenge
Build a working product on one graph problem. FalkorDB has to power a core feature of it, not decorate one.
The problem
Vector RAG breaks down on multi-step reasoning and on entities spread across documents. That is where models hallucinate.
The shift
Raw data structured as nodes and edges gives the model an explicit map to reason over.
The brief
Combine knowledge graphs, LLMs, and autonomous agents into something fast, accurate, and explainable.
Work the graph has to do
Prize pool
$10,000
Split across three tracks. The final prize breakdown and the team prize policy are announced before the hackathon begins.
Track 01
Build a product that uses connected data to investigate fraud, identity, claims, transactions, or other linked activity, and can show the reader how it reached its answer.
Key focus
Prize details coming soon
Track 02
Build autonomous AI agents, or a swarm of them, that use FalkorDB as a persistent cognitive memory layer: episodic, semantic, and procedural memory that survives across sessions.
Key focus
Prize details coming soon
Track 03
Build production-grade, reliable data pipelines on the official open-source GraphRAG SDK.
Key focus
Prize details coming soon
02 / Why graphs
These are hard because the answer is spread across several relationships rather than sitting in any one record.
As rows
logins
| account | device |
|---|---|
| acct_18 | dev_a3 |
| acct_42 | dev_a3 |
| acct_77 | dev_b1 |
| acct_91 | dev_a3 |
As a graph
same four rows
Pull a thread
tap a node
Three accounts land on one device. Follow one step further and both devices were seen at the same address, so the ring is larger than the table said it was.
The example is illustrative, but the shape is not. Every question worth asking on this page has the same one: the answer is spread across several relationships, and a graph database stores those relationships directly rather than rebuilding them with joins on every query.
03 / About the sponsor
An open-source property graph database with the GraphRAG SDK, Cypher, sub-millisecond traversals, and clients for the stack you already use.
An open-source property graph database built for AI workloads, with support for the Cypher query language and traversals fast enough to sit inside an agent loop. Run it locally with Docker, or use FalkorDB Cloud and skip the setup entirely.
FalkorDB Browser is the visual side of it: create graphs, run queries, and see how the records you loaded actually connect.
Official clients
Cypher
who shares a device
MATCH (a:Account)-[:USED]->(d:Device)<-[:USED]-(b:Account)
WHERE a.id < b.id
RETURN d.id AS device, collect(a.id) + collect(b.id) AS accounts
ORDER BY size(accounts) DESC
LIMIT 10The middle of the first line is a drawing of the thing you are looking for. Two accounts, one device, arrows either side. That pattern is the query.
Point it at your raw documents and it builds the graph: the ontology is generated for you, and retrieval runs across vector, graph, and keyword at once.
Sub-millisecond traversals make this a memory layer an agent can read and write on every step, and still find there on the next session.
Create nodes and relationships, match graph patterns, follow paths, filter results, and return the data your application needs.
The traversals you would otherwise write yourself, already in the database.
Use FalkorDB Browser to run queries, inspect the data model, review results, and visualise how records connect.
Connect through the official clients, or through the AI framework you are already writing against.
04 / FAQ
The hackathon is open to developers worldwide. You can participate solo or with a team of up to four people.
No. The getting started guide and the kickoff session cover the basics you need to begin.
Yes. FalkorDB must be the primary graph database, and it must support a central part of the product.
Registration happens through the Google Form linked from the Register now button on this page. There is no account to create here.
Yes. You may use any language, framework, frontend, or deployment platform that works with your project.
You may reuse general libraries, templates, and infrastructure. The submission and its main FalkorDB implementation must be new work completed during the event.
Yes. You remain responsible for the code, architecture, security, and correctness of the project, and you must be able to explain what was built and why.
A live deployment is preferred. Projects that cannot be deployed must include complete local setup instructions and a working demo video.