Hillock v0.7: replacing vector DBs and LLM extraction with SQLite, Hebbian links, HDC gating
Equivalent-Flan-1590 · reddit · 2026-09-28
The developer of Hillock (AGPL-3.0) wrote a detailed post on the architectural decisions behind v0.7 of the neuro-symbolic memory engine, explaining why it abandons dense vector stores for local document memory.
Bottlenecks of standard RAG:
- Generative-LLM ingestion is quadratic and bottlenecked by token generation rates;
- Cosine similarity over dense embeddings lacks a strict "I don't know" threshold (out-of-domain rejection), causing downstream context poisoning.
Hillock's approach:
- TALON triple extraction: bypasses generative LLMs entirely — document-level coreference (fastcoref), predicate routing (MiniLM), zero-shot relation extraction (GLiREL) produce SPO triples via O(1) matrix classifications at <300MB VRAM;
- Plastic graph persistence: facts in SQLite (WAL), with local Hebbian synaptic weights instead of expensive graph-wide PageRank;
- HDC/VSA gating: entities, relations and n-grams projected into D=10,000 bipolar hypervector space, scored with ColBERT-style late-interaction MaxSim; below-threshold queries are terminated before the generative SLM is called.
v0.7 adds source provenance tables, proactive Hebbian prompt injection and human-in-the-loop disambiguation. The author requests technical critique of the HDC implementation.
Related event: Hillock: Open-Source Neuro-Symbolic Memory Engine Ditches Vector DBs(2 posts)→
More from coding & agent
- Community Builds Timeline of 17 AI Agent Incidents, Linking Official OpenAI and Anthropic Reports — gleech · 2026-09-28
- 5 AI Agent Security Risks: More Autonomy Demands Tighter Controls — goyalshaliniuk · 2026-09-28
- Dev blasts Google AX's agent sandbox DX, offers 3-line alternative with Celesto — aniketmaurya · 2026-09-28
- Dev says $100-200 plans removed token limits; review cycles and human bandwidth are now the bottleneck — ezshine · 2026-09-28
- LangWatch benches open-source Jev alternatives on Jev's own benchmark — _rchaves_ · 2026-09-28
- Genomics MCP v0.1.0: 23 open-source tools let AI agents fetch genomic reads, variants and signal — Fair-Rain3366 · 2026-09-28