Crate engram 

Engram — durable memory layer for AI agents.

Provides a temporal knowledge graph, semantic search, and MCP-native tools for agents running on the JamJet runtime. Memory is scoped, versioned, and queryable across time — enabling agents to reason over what they knew, when.

Re-exports

pub use consolidation::ConsolidationConfig;

pub use consolidation::ConsolidationEngine;

pub use consolidation::ConsolidationOp;

pub use consolidation::ConsolidationResult;

pub use context::CharTokenEstimator;

pub use context::ContextBlock;

pub use context::ContextBuilder;

pub use context::ContextConfig;

pub use context::OutputFormat;

pub use context::TokenEstimator;

pub use embedding::EmbeddingProvider;

pub use embedding_ollama::OllamaEmbeddingProvider;

pub use extract::ExtractedFact;

pub use extract::ExtractionConfig;

pub use extract::ExtractionResult;

pub use extract::Message;

pub use fact::Entity;

pub use fact::EntityId;

pub use fact::Fact;

pub use fact::FactFilter;

pub use fact::FactId;

pub use fact::FactPatch;

pub use fact::MemoryTier;

pub use fact::Relationship;

pub use fact::RelationshipId;

pub use fact::SubGraph;

pub use graph::GraphStore;

pub use graph_postgres::PostgresGraphStore;

pub use graph_sqlite::SqliteGraphStore;

pub use llm::LlmClient;

pub use llm_anthropic::AnthropicLlmClient;

pub use llm_command::CommandLlmClient;

pub use llm_google::GoogleLlmClient;

pub use llm_ollama::OllamaLlmClient;

pub use llm_openai::OpenAiLlmClient;

pub use memory::Memory;

pub use message::ChatMessage;

pub use message::MessageId;

pub use message::MessageStore;

pub use message_postgres::PostgresMessageStore;

pub use message_sqlite::SqliteMessageStore;

pub use pipeline::ExtractionPipeline;

pub use scope::Scope;

pub use store::FactStore;

pub use store::MemoryError;

pub use store::StoreStats;

pub use store_postgres::PostgresFactStore;

pub use store_sqlite::SqliteFactStore;

pub use vector::VectorFilter;

pub use vector::VectorMatch;

pub use vector::VectorStore;

pub use vector_embedded::EmbeddedVectorStore;

Modules

conflict

Conflict detection — Stage 2 of the extraction pipeline.

consolidation

Consolidation engine — background memory maintenance.

context

Context assembly — token-budgeted context building for LLM prompts.

embedding

EmbeddingProvider trait — pluggable text-embedding interface for Engram.

embedding_ollama

Ollama-backed EmbeddingProvider implementation.

extract

Extraction pipeline — types and configuration.

fact

Core domain types for the Engram memory layer.

graph

GraphStore trait — the storage interface for the entity-relationship graph.

graph_postgres

PostgreSQL-backed GraphStore implementation.

graph_sqlite

SQLite-backed GraphStore implementation.

llm

LlmClient trait — lightweight LLM interface for extraction.

llm_anthropic

Anthropic Claude LLM client — Messages API.

llm_command

CommandLlmClient — the shell-out extensibility escape hatch.

llm_google

Google Gemini LLM client — generateContent API.

llm_ollama

Ollama LLM client — free, local chat completion.

llm_openai

OpenAI-compatible LLM client — chat completions with native JSON mode.

llm_util

Shared utilities for LLM clients.

memory

Memory — the primary public API for the Engram memory layer.

message

ChatMessage type and MessageStore trait — chat message persistence.

message_postgres

PostgreSQL-backed MessageStore implementation.

message_sqlite

SQLite-backed MessageStore implementation.

pipeline

Extraction pipeline — LLM-based fact, entity, and relationship extraction.

retrieve

Hybrid retrieval — merges vector, keyword, and graph search results.

scope

Scope — hierarchical identity context for memory isolation.

store

FactStore trait — the primary storage interface for Engram.

store_postgres

PostgreSQL-backed FactStore implementation.

store_sqlite

SQLite-backed FactStore implementation.

vector

VectorStore trait — semantic similarity search interface for Engram.

vector_embedded

EmbeddedVectorStore — in-process brute-force cosine-similarity store.