prototype
RootGraph
Graph-based context for AI agents running fully offline on a small Qwen3 model served with vLLM. Built at Graph Hacks (FalkorDB × WeMakeDevs), October 2026.
Problem
Small local models lose track of context quickly. Long prompts are slow and expensive, and plain vector search misses how facts relate to each other.
Approach
- Store the agent’s working context as a graph in FalkorDB, so related facts and entities are linked rather than flattened.
- Retrieve a focused subgraph for each step instead of stuffing the whole history into the prompt.
- Run the model fully offline: a small Qwen3 model served with vLLM.
Context
Built during the Graph Hacks hackathon by FalkorDB and WeMakeDevs, October 2026.
Stack
FalkorDB, vLLM, Qwen3, Python.