Memory Types & Subtypes Reference

Cognitive Types

Type	Enum Value	Description	Retention	Typical Use
Episodic	episodic	Specific events/interactions	Decays over time	Event logs, interaction history
Semantic	semantic	Facts, concepts, relationships	Permanent until modified	Preferences, decisions, knowledge
Procedural	procedural	How-to knowledge	Permanent	Solutions, patterns, workflows
Working	working	Current task context	Session-scoped (TTL)	Active task tracking

Python

from memorylayer import MemoryType

MemoryType.EPISODIC    # "episodic"
MemoryType.SEMANTIC    # "semantic"
MemoryType.PROCEDURAL  # "procedural"
MemoryType.WORKING     # "working"

TypeScript

import { MemoryType } from "@scitrera/memorylayer-sdk";

MemoryType.EPISODIC    // "episodic"
MemoryType.SEMANTIC    // "semantic"
MemoryType.PROCEDURAL  // "procedural"
MemoryType.WORKING     // "working"

Domain Subtypes

Subtype	Enum Value	Description
Solution	solution	Working fixes to problems
Problem	problem	Issues encountered
CodePattern	code_pattern	Reusable code patterns
Fix	fix	Bug fixes with context
Error	error	Error patterns and resolutions
Workflow	workflow	Process knowledge
Preference	preference	User or project preferences
Decision	decision	Architectural decisions
Profile	profile	Person or entity profiles
Entity	entity	Named entities (people, places, things)
Event	event	Significant events or milestones
Directive	directive	User instructions and constraints
Inference	inference	Derived/inferred facts attributed to a reasoning step
Fact	fact	Atomic facts produced by fact decomposition; feeds the fact retrieval channel

fact is server-side only. The server accepts and emits fact, but it is not yet in the MemorySubtype enum shipped by the Python and TypeScript SDKs. Pass it as the string "fact" from the SDKs until the enums catch up.

Python

from memorylayer import MemorySubtype

MemorySubtype.SOLUTION       # "solution"
MemorySubtype.PROBLEM        # "problem"
MemorySubtype.CODE_PATTERN   # "code_pattern"
MemorySubtype.FIX            # "fix"
MemorySubtype.ERROR          # "error"
MemorySubtype.WORKFLOW       # "workflow"
MemorySubtype.PREFERENCE     # "preference"
MemorySubtype.DECISION       # "decision"
MemorySubtype.PROFILE        # "profile"
MemorySubtype.ENTITY         # "entity"
MemorySubtype.EVENT          # "event"
MemorySubtype.DIRECTIVE      # "directive"
MemorySubtype.INFERENCE      # "inference"

TypeScript

import { MemorySubtype } from "@scitrera/memorylayer-sdk";

MemorySubtype.SOLUTION       // "solution"
MemorySubtype.PROBLEM        // "problem"
MemorySubtype.CODE_PATTERN   // "code_pattern"
MemorySubtype.FIX            // "fix"
MemorySubtype.ERROR          // "error"
MemorySubtype.WORKFLOW       // "workflow"
MemorySubtype.PREFERENCE     // "preference"
MemorySubtype.DECISION       // "decision"
MemorySubtype.DIRECTIVE      // "directive"
MemorySubtype.PROFILE        // "profile"
MemorySubtype.ENTITY         // "entity"
MemorySubtype.EVENT          // "event"
MemorySubtype.INFERENCE      // "inference"

Recall Modes

Mode	Enum Value	Description	Status
RAG	rag	Hybrid retrieval: dense vector + keyword (BM25) fused via RRF, with graph traversal and intent-gated arms	Active (default)
Agentic	agentic	Iterative, LLM-driven retrieval loop over the same arms	Server-side; not in the SDK enums
LLM	llm	LLM-powered semantic search	Deprecated
Hybrid	hybrid	Combination of RAG and LLM	Deprecated

Note: LLM and Hybrid modes are being phased out. Use RAG mode for recall and the Context Environment with RLM for LLM-powered analysis over memories.

Importance Scoring Guide

Score	Level	Use For
0.9–1.0	Critical	Security fixes, breaking changes, core architecture
0.7–0.8	High	Bug fixes, API changes, important patterns
0.5–0.6	Standard	Preferences, general knowledge
0.3–0.4	Low	Observations, temporary workarounds
0.0–0.2	Ephemeral	Debugging traces, session context