Analyze a message and determine if it contains memorable information. Returns a standardized MemoryEntry JSON with type, tier, confidence, and suggested_action. CarryMem is a CarryMem memory system with optional storage — it tells you WHAT to remember, and can optionally store it too.

Return CarryMem's complete classification schema definition including 7 memory types, 4 storage tiers, confidence thresholds, and downstream mapping tables.

Batch classify multiple messages, each returning an independent MemoryEntry.

Classify a message AND store it if worth remembering. One-step operation: classify → store → return. Requires storage adapter to be configured.

Retrieve stored memories. Supports filtering by type, tier, and confidence. Supports full-text search. Requires storage adapter.

Delete a stored memory by ID. Requires storage adapter.

Index an Obsidian vault or knowledge base for full-text search. Scans Markdown files, extracts YAML frontmatter tags and wiki-links, builds FTS5 index. Requires knowledge adapter (ObsidianAdapter).

Search knowledge base (e.g., Obsidian vault) using full-text search. Returns matching notes with title, content preview, tags, and wiki-links. Requires knowledge adapter.

Unified retrieval across both memories (SQLite) and knowledge base (Obsidian). Returns results from both sources with priority: memories first, then knowledge. Requires at least one adapter configured.

Let the user proactively tell the AI about themselves. User declarations are classified by the engine but always stored with confidence=1.0 and source_layer='declaration'. Active declaration + passive classification = complete memory coverage.

Get a structured summary of what the AI remembers about the user. Returns highlights (top preferences, decisions, corrections), statistics (by type, by tier, avg confidence), and a human-readable summary. Lets users see and audit what AI remembers.

Generate a system prompt with user memories and knowledge base context injected. The prompt follows the 'memory-first' retrieval priority: User Memories > Knowledge Base > General Knowledge. Use this to inject CarryMem context into any AI agent's system prompt.

Request the host AI to summarize conversation content, then store the summary as a session_summary memory. This implements the 'borrow host LLM' pattern: CarryMem returns the content that needs summarizing, the host AI generates a concise summary focusing on user preferences, decisions, and key facts, then calls classify_and_remember or declare_preference to store it. No external LLM API key needed.

Run memory consolidation: deduplicate similar memories, apply time-based decay, clean up low-value entries, and detect patterns for rule promotion. Preferences are always preserved. P0 handles dedup+decay, P1 detects repeated patterns and generates rule candidates. Run periodically (e.g., daily) to keep memory store healthy. Use dry_run=true first to preview changes.

Schedule periodic memory consolidation (dedup, decay, cleanup) at a fixed interval

Stop the scheduled periodic memory consolidation

Add a behavioral rule to CarryMem's rule engine. Rules guide AI behavior for specific topics. Use 'company' scope for organization-mandated rules (highest priority), 'negotiated' for team-adapted rules, or 'personal' for individual preferences (lowest priority).

List all rules in CarryMem's rule engine, optionally filtered by scope or trigger topic.

Match rules against a scene/topic and return applicable rules with scores. Use this to find which rules apply to a given context before generating a response.

Generate a formatted rules section for injection into AI prompts. Returns structured text with applicable rules for a given context, including scope labels and priority markers.

View all your saved rules in a readable summary format. Shows rule triggers, actions, scope, type, and override status. Use this to review what CarryMem remembers about your preferences and behavioral rules.

Delete a rule by its ID. Use my_rules first to find the rule ID you want to remove. Returns confirmation with the deleted rule's details.

Analyze your stored memories and suggest rule candidates based on detected patterns. Useful for discovering preferences you've expressed multiple times that could become formal rules.

Run the full promotion pipeline: analyze memories, detect patterns, generate rule candidates, and optionally auto-accept them. Use this to convert accumulated preferences into active rules.

Update an existing rule's trigger, action, scope, or type. Use my_rules first to find the rule ID. Returns the updated rule details.

Get a complete view of your CarryMem identity: memory statistics, rule summary, recent activity, and preference distribution. Use this to understand what CarryMem knows about you.

First-time user onboarding. Returns a welcome message and asks key preference questions to initialize your CarryMem profile. Call this when a new user starts their first conversation.

Check CarryMem system health. Returns adapter health, audit logger stats, memory count, and uptime. Lightweight check that does not start any HTTP service — use this from MCP clients to verify CarryMem is operational.

Multi-hop graph traversal from an entity. Performs BFS traversal of the knowledge graph starting from the given entity, collecting all connected entities and memories within max_hops hops. Requires storage adapter with graph capability.

Find the shortest path between two entities in the knowledge graph using bidirectional BFS. Returns the path as a list of entity texts from source to destination. Useful for understanding how concepts are connected. Requires storage adapter with graph capability.

Compute the graph impact of a memory. Returns the number of entities linked to the memory, the number of relations it evidences, whether it spans multiple namespaces, and an impact_score (entity_count0.4 + relation_count0.4 + cross_namespace*0.2). Requires storage adapter with graph capability.