Skip to main contentConcepts
Membria CE is built around a small set of runtime primitives. This page summarizes them and points to the deeper docs.
Core concepts
Decision Surface (DS)
The Decision Surface is the user-facing layer that shows open loops, drift, and outcomes. It is derived from DBB records, not raw chat.
See Client overview and Runtime diagrams.
Decision Black Box (DBB)
DBB is the capture engine that detects decisions and records structured evidence with confidence scores.
See DBB and LoRA runtime and Client runtime RFC.
ThoughtUnits and GraphRAG
ThoughtUnits are normalized fragments that feed the graph and vector layers. GraphRAG uses the temporal graph + vectors to produce explainable retrieval.
See Client runtime RFC and Runtime diagrams.
Cache-augmented generation (CAG)
Verified answers are cached locally and in the KCG so responses improve over time without re-computation.
See Decentralized backend and KCG and gateways.
Escalation and Council governance
Escalation routes queries to stronger models only when local confidence fails. The Council output is verified and written back with provenance.
See Decentralized backend, KCG and gateways, and Client runtime RFC.
LoRA adapters and SkillForge
LoRA adapters provide scoped expertise without changing the base model or rewriting history.
See DBB and LoRA runtime.
Where to go next
