Module memory 

Modules

agent_scoped

Runtime memory wrapper bound to one agent.

agent_scoped_markdown

Cross-agent path-walk variant for Markdown-backed agents.

audit

Audit trail for memory operations.

backend

chunker

cli

conflict

Conflict resolution for memory entries.

consolidation

LLM-driven memory consolidation.

decay

embeddings

hygiene

importance

Heuristic importance scorer for non-LLM paths.

knowledge_graph

Knowledge graph for capturing, organizing, and reusing expertise.

lucid

markdown

none

policy

Policy engine for memory operations.

qdrant

response_cache

Response cache — avoid burning tokens on repeated prompts.

retrieval

Multi-stage retrieval pipeline.

snapshot

Memory snapshot — export/import core memories as human-readable Markdown.

sqlite

traits

vector

Structs

AgentScopedMarkdownMemory

Composed Markdown memory for one agent: own backend plus the resolved peer set. Stores write only to the bound agent; recalls union across own + peers with per-row alias attribution.

AgentScopedMemory

A Memory impl that scopes every read and write to a bound agent’s UUID + a resolved cross-agent allowlist.

AuditedMemory

Decorator that wraps a Memory backend with audit logging.

ExportFilter

Filter criteria for bulk memory export (GDPR Art. 20 data portability).

LucidMemory

MarkdownMemory

Markdown-based memory — plain files as source of truth

MarkdownPeer

Resolved Markdown-backed peer entry: the sibling agent’s alias plus a MarkdownMemory pointed at that sibling’s workspace dir.

MemoryBackendProfile

MemoryEntry

A single memory entry

NoneMemory

Explicit no-op memory backend.

PolicyEnforcer

Policy enforcer that validates memory operations.

ProceduralMessage

A single message in a conversation trace for procedural memory.

QdrantMemory

Qdrant vector database memory backend.

ResponseCache

Two-tier response cache: in-memory LRU (hot) + SQLite (warm).

RetrievalConfig

Multi-stage retrieval pipeline configuration.

RetrievalPipeline

Multi-stage retrieval pipeline wrapping a Memory backend.

SqliteMemory

SQLite-backed persistent memory — the brain

Enums

MemoryBackendKind

MemoryCategory

Memory categories for organization

Constants

MEMORY_CONTEXT_CLOSE

Closing delimiter for recalled memory injected into provider context.

MEMORY_CONTEXT_OPEN

Opening delimiter for recalled memory injected into provider context.

Traits

Memory

Core memory trait — implement for any persistence backend

Functions

backend_kind_from_dotted

Extract the backend kind from a V3 dotted reference (<kind>.<alias>). Bare names ("sqlite") are returned as-is. Returned lowercase.

classify_memory_backend

create_memory

Factory: create the right memory backend from config

create_memory_for_agent

Build the per-agent memory wrapper for agent_alias.

create_memory_for_migration

create_memory_with_storage_and_routes

Factory: create memory with a resolved active storage backend and embedding routes.

create_response_cache

Factory: create an optional response cache from config.

default_memory_backend_key

is_assistant_autosave_key

Legacy auto-save key used for model-authored assistant summaries. These entries are treated as untrusted context and should not be re-injected.

is_recent_recall_query

Returns true when a recall query should be interpreted as recent/time-only recall.

is_user_autosave_key

Auto-save key used for raw user messages captured per-turn. Re-injecting these into build_context causes exponential bloat: each recalled entry contains prior generations’ context verbatim, growing unboundedly. Consolidated knowledge is already promoted to Core/Daily entries.

memory_backend_profile

normalize_recent_recall_query

Normalizes recent/time-only recall queries to the backend-neutral empty query.

selectable_memory_backends

should_skip_autosave_content

Filter known synthetic autosave noise patterns that should not be persisted as user conversation memories.