How roandejager/hillock works
Hillock is a local neuro-symbolic memory engine for extracting document facts and answering questions against them. Users can interact through the terminal console, an OpenAI-compatible API, or the Python engine. The main workflow ingests text or PDFs through TALON’s resolution, predicate-routing, and relation-extraction stages, then records symbolic and vector/synaptic memory. On queries, the engine selects evidence and applies a hard answerability gate before optional Ollama rendering. The graph reflects the documented workflow and shown imports/calls; internal engine details not visible in excerpts are kept at subsystem level.
Main entry points: engine.py Hillock engine · ingestor.py Document ingestor · talon_engine.py TALON orchestrator · main.py Terminal console
Components 14
User interfaces
- Terminal console
main.py - OpenAI API
api.py
Fact ingestion
- Document ingestor
ingestor.py - TALON orchestrator
talon_engine.py - Coreference resolver
talon_engine.py - Predicate router
talon_engine.py - Relation extractor
talon_engine.py
Memory systems
- SQLite knowledge graph
database.py - Hebbian associations
plasticity.py - HDC reservoir
reservoir.py
Query and response
- Hillock engine
engine.py - Evidence selection gate
engine.py
Other components
- User
- Ollama service
Connections 16
- User → Terminal console: uses
README.md - User → OpenAI API: connects
README.md - Terminal console → Hillock engine: submits queries
main.py - Terminal console → Document ingestor: starts ingestion
README.md - OpenAI API → Hillock engine: executes turn
api.py - Document ingestor → TALON orchestrator: extracts triples
ingestor.py - TALON orchestrator → Coreference resolver: resolves references
talon_engine.py - TALON orchestrator → Predicate router: routes predicates
talon_engine.py - TALON orchestrator → Relation extractor: extracts relations
talon_engine.py - Document ingestor → Hillock engine: resolves entities
ingestor.py - Document ingestor → HDC reservoir: updates HDC state
ingestor.py - Hillock engine → SQLite knowledge graph: reads facts
engine.py - Hillock engine → Hebbian associations: uses associations
engine.py - Hillock engine → HDC reservoir: uses HDC memory
engine.py - Hillock engine → Evidence selection gate: selects evidence
engine.py - Evidence selection gate → Ollama service: renders answer
engine.py