mnemara-core

mnemara-core provides the product-neutral memory domain model, query types, scoring configuration, evaluation helpers, portable package types, and async store traits that the rest of the Mnemara workspace builds on.

Choose this crate when you need the typed memory model and retrieval or maintenance contracts, but not a specific storage backend or transport surface.

Install

Add the crate to your Rust project with:

cargo add mnemara-core

What it contains

MemoryRecord, MemoryScope, MemoryRecordKind, MemoryQualityState, MemoryHistoricalState, and MemoryTrustLevel
episodic context types, continuity state, salience, recurrence, duration, boundary cues, and lineage links
RecallQuery, RecallFilters, RecallResult, recall explanations, and continuity-aware planning trace types
BatchUpsertRequest, UpsertRequest, DeleteRequest, ArchiveRequest, SuppressRequest, RecoverRequest, and admin operation reports
SnapshotManifest, portable export or import package types, integrity or repair reports, and maintenance stats
EngineConfig, scorer kinds, scoring and policy profiles, planning profiles, embedding-provider configuration, and evaluation helpers
MemoryStore, the async trait implemented by the file and sled backends

Minimal example

use mnemara_core::{
    EngineConfig, RecallFilters, RecallHistoricalMode, RecallQuery, RecallScoringProfile,
    RecallTemporalOrder,
};

let mut config = EngineConfig::default();
config.recall_scoring_profile = RecallScoringProfile::Balanced;

let query = RecallQuery {
    query_text: "reconnect storm mitigation".to_string(),
    filters: RecallFilters {
        unresolved_only: true,
        historical_mode: RecallHistoricalMode::IncludeHistorical,
        temporal_order: RecallTemporalOrder::NewestFirst,
        ..Default::default()
    },
    ..Default::default()
};

Notes

additive schema evolution keeps missing episodic or lifecycle fields backward compatible for existing JSON records and portable packages
embedded callers can keep record-only usage, or opt into episodic recall, planner traces, lineage-aware retrieval, and lifecycle controls as needed
custom semantic providers can integrate through the public embedding seam without changing the domain model

Related crates

mnemara for the facade crate
mnemara-store-file for the file-backed implementation
mnemara-store-sled for the embedded sled-backed implementation
mnemara-protocol for protobuf and gRPC surface types
mnemara-server for the daemon crate and reusable service surfaces