Expand description
Memgine — graph-based memory engine for Common Agent Runtime.
Memory is a graph. Nodes are facts, skills, conclusions, conversations, identity, environment. Edges: Supersedes, DependsOn, RelatedTo, Triggers, TemporalNext, CitesPremise.
Skills are learned procedures — stored as graph nodes with trigger edges, matched via spreading activation, tracked for success/failure.
Conclusions are derived knowledge citing premise facts — automatically invalidated when premises change (cascade invalidation).
Re-exports§
pub use compaction::CompactionResult;pub use compaction::ConversationCompactionReport;pub use config::ContextMode;pub use config::MemgineConfig;pub use conversation_store::ConversationStore;pub use conversation_store::StoredTurn;pub use distill::DistilledSkill;pub use distill::TraceEvent;pub use engine::ConsolidationReport;pub use engine::DomainStats;pub use engine::MemgineEngine;pub use graph::detect_content_type;pub use graph::stem;pub use graph::synonym_expand;pub use graph::tokenize_for_content;pub use graph::CodeLanguage;pub use graph::ContentType;pub use graph::EdgeKind;pub use graph::FactMetadata;pub use graph::MemEdge;pub use graph::MemKind;pub use graph::MemNode;pub use graph::MemoryGraph;pub use graph::Partition;pub use graph::Provenance;pub use graph::RetrievalHit;pub use graph::SkillMeta;pub use graph::SkillOutcome;pub use graph::SkillScope;pub use graph::SkillStats;pub use graph::SkillTrigger;pub use project::ConfigOverrides;pub use project::DotCarProject;pub use project::KnowledgeEntry;pub use project::PolicyRule;pub use reflection::ReflectionInsight;pub use reflection::ReflectionReport;pub use trajectory::Trajectory;pub use trajectory::TrajectoryOutcome;pub use trajectory::TrajectoryStats;pub use trajectory::TrajectoryStore;
Modules§
- compaction
- Conversation compaction — embedding-aware, graph-informed progressive summarization.
- confidence
- Graduated confidence scoring for facts.
- config
- Configuration for the memgine engine.
- constraint_
checker - Deterministic constraint pre-computation.
- conversation_
store - Persistent conversation storage with bounded write-through.
- distill
- Trajectory distillation — extract skills from execution traces.
- engine
- Memgine engine — graph-based memory orchestrator.
- graph
- The memory graph — nodes are memory entries, edges are relationships.
- project
- .car/ project directory — team-shareable configuration and knowledge.
- query_
classifier - Query complexity classifier for adaptive context assembly.
- reflection
- Self-reflection — learns from conversation patterns.
- trajectory
- Trajectory storage — persists raw (proposal, trace, outcome) tuples for offline analysis, replay, and future RL-style learning.
Structs§
- Node
Index - Node identifier.