Memory Categories

Understanding REMBR's 12 memory categories and when to use each type

Why Categories?

Categories help the AI automatically retrieve the most relevant context. When you search for "preferences", REMBR knows to prioritize memories in the preferences category. The AI also auto-categorizes memories during storage based on content.

Original Categories (8)

General-purpose categories for everyday AI assistant use

`facts`

Concrete information and data points

EXAMPLES:

•"The API rate limit is 1000 requests/hour"

•"User authentication uses JWT with RS256"

•"Database runs PostgreSQL 16"

USE FOR:

System properties, configuration values, technical specifications

`preferences`

User preferences and settings

EXAMPLES:

•"User prefers TypeScript over JavaScript"

•"Team uses tabs (width: 2)"

•"Favors functional programming patterns"

USE FOR:

Coding style, tool preferences, workflow choices

`conversations`

Conversation history and context

EXAMPLES:

•"Discussed implementing OAuth 2.0 on Jan 15"

•"User asked about database migration"

•"Clarified requirements for async tasks"

USE FOR:

Session context, discussion summaries, Q&A history

`projects`

Project-specific information

EXAMPLES:

•"Project uses Next.js 14 App Router"

•"Main repo: github.com/company/project"

•"Deploy via Kubernetes on ARM64"

USE FOR:

Project metadata, technology stack, deployment details

`learning`

Knowledge and insights learned

EXAMPLES:

•"pgvector HNSW is faster than IVF for <100k vectors"

•"React Server Components reduce bundle size"

•"Redis improves auth performance"

USE FOR:

Discoveries, lessons learned, best practices identified

`goals`

Objectives and targets

EXAMPLES:

•"Reduce API latency below 100ms"

•"Deploy multi-region by Q2"

•"Implement real-time notifications"

USE FOR:

Project goals, performance targets, milestones

`context`

Situational context and state

EXAMPLES:

•"Currently working on authentication refactor"

•"Blocked on database migration approval"

•"Testing new search algorithm"

USE FOR:

Current state, blockers, work-in-progress

`reminders`

Future actions and reminders

EXAMPLES:

•"TODO: Update API docs after auth changes"

•"Test edge cases for null values"

•"Benchmark new caching layer"

USE FOR:

Action items, follow-ups, scheduled tasks

RLM-Optimized Categories (4)

Specialized categories for Recursive Language Models and agent systems

`patterns`

Code patterns, architectural patterns, best practices

EXAMPLES:

•"Repository pattern used for data access"

•"Dependency injection via constructor"

•"Error handling: try-catch with custom types"

USE FOR:

Design patterns, code conventions, architectural decisions

RLM Benefit: Perfect for RLM agents analyzing codebases to remember discovered patterns

`decisions`

Technical decisions, trade-offs, architectural choices

EXAMPLES:

•"Chose PostgreSQL over MongoDB for ACID"

•"Using Prisma ORM for type safety"

•"Rejected microservices - monolith simpler"

USE FOR:

Architecture decisions, technology choices, trade-off analyses

RLM Benefit: Critical for RLM decision-making: 'Why was this approach chosen?'

`workflows`

Process flows, deployment procedures, development workflows

EXAMPLES:

•"Deploy: build → test → push Docker → kubectl rollout"

•"PR review: linting → tests → 2 approvals"

•"Hotfix: branch from main → test → merge"

USE FOR:

CI/CD pipelines, development processes, operational procedures

RLM Benefit: Helps RLM agents understand and follow established processes

`insights`

Analytical findings, performance insights, optimization opportunities

EXAMPLES:

•"95th percentile latency spikes at 1000+ concurrent users"

•"Database connection pool exhausted during peak"

•"Caching reduced API calls by 60%"

USE FOR:

Performance analysis, bottleneck identification, optimization results

RLM Benefit: Enables data-driven decision making for recursive agents

How the AI Uses Categories

1. Automatic Categorization

When you store a memory, the AI analyzes the content and automatically suggests the most appropriate category. You can override this if needed.

User: "Remember I prefer TypeScript"
→ Auto-categorized as preferences

2. Intelligent Search Filtering

When searching, the AI can filter by category or boost relevance for certain categories based on query intent.

Query: "What's my coding style?"
→ Searches preferences category first

3. Context Workspaces

RLM agents can create context workspaces filtered by category - e.g., a "decisions" context for architectural choices only.

4. Analytics & Insights

Category distribution helps you understand what type of knowledge your system has accumulated and identify gaps.

Category Best Practices

✓

Use Specific Categories

Choose decisions over factswhen storing why a choice was made, not just what was chosen.

✓

Trust Auto-Categorization

The AI is pretty good at picking categories. Only override if clearly wrong.

✓

Use RLM Categories for Agents

When building recursive agent systems, leverage patterns, decisions, workflows, and insights for better context management.

✗

Don't Over-Think It

Categories are hints, not rigid rules. Search works across all categories anyway.

Learn More About MCP Tools

Explore all 98 MCP Tools for working with memories