Show HN: Hillock – Local neuro-symbolic memory engine in <1.2GB VRAM (github.com)

Hillock, a novel local neuro-symbolic memory engine, has been introduced as a solution tailored for edge hardware, boasting a minimal VRAM requirement of under 1.2 GB. Unlike traditional models that rely on expansive vector databases and generative LLMs, Hillock employs a deterministic similarity gate to filter unanswerable questions before they reach the LLM stage, ensuring more reliable responses. By utilizing a lightweight architecture that combines relational knowledge graphs, Hebbian synaptic memory, and high-dimensional vector symbolic architectures, Hillock can ingest documents in approximately 5 seconds and operate entirely offline, negating cloud dependencies. This development is significant for the AI/ML community as it offers an efficient and resource-conserving alternative to conventional large language model-based processing, which often suffers from hallucinations and requires extensive computational resources. The engine's performance metrics highlight its capability to handle structured outputs deterministically with a 100% pipeline completion rate, while maintaining a significantly lower ingestion latency compared to standard methods. The latest version, v0.5.0, enhances user interaction with quickstart launchers and an expanded CLI for real-time monitoring, further solidifying Hillock's role in enabling efficient AI applications on edge devices.