Module mnemo 

Native memory coprocessor for persistent recall across sessions.

Inspired by Mnemo Cortex, this module provides:

• SQLite + FTS5 storage for messages and summaries
• Background compaction via LLM summarization
• Context injection at agent bootstrap
• DAG-based summary lineage for expansion

Trait Architecture

Follows existing RustyClaw patterns:

• MemoryStore — storage abstraction (SQLite, in-memory, remote)
• Summarizer — summarization backend (LLM, deterministic)

Structs

CompactionStats

Statistics from a compaction pass.

DeterministicSummarizer

Deterministic fallback summarizer (truncation-based).

LlmSummarizer

LLM-backed summarizer using provider infrastructure.

MemoryEntry

A single memory entry (message or summary).

MemoryHit

Search result with relevance score.

MnemoConfig

Configuration for the memory coprocessor.

SqliteMemoryStore

SQLite-backed memory store with FTS5 support.

SummarizationConfig

Summarization backend configuration.

Enums

SummaryKind

Kind of summary being generated.

Traits

MemoryStore

Core memory storage trait.

Summarizer

Summarization backend for memory compaction.

Functions

create_memory_store

Create a shared memory store from configuration.

create_summarizer

Create a summarizer based on configuration.

estimate_tokens

Estimate token count for a string (rough heuristic: ~4 chars per token).

generate_context_md

Generate context markdown for bootstrap injection.

run_compaction

Run compaction in a background loop.

Type Aliases

SharedMemoryStore

Shared mnemo store for concurrent access.