Memory Types

Every memory in Brain Memory is classified into one of eight types. Each type has a different base strength and decay rate, reflecting its typical importance and how long it should persist.

Type Overview

Detailed Descriptions

Insight (0.90)

Insights are deep realizations or patterns you have discovered. They start with the highest base strength because they represent distilled wisdom that is hard to re-derive.

Examples:

• "React Server Components are most effective when data fetching is colocated with the component that consumes it"
• "Our deploy failures correlate with PRs that skip the staging environment"
• "The team moves faster when architecture decisions are documented in ADRs"

Decision (0.85)

Decisions capture choices made and the reasoning behind them. These are critical for maintaining consistency and avoiding re-litigating settled questions.

Examples:

• "Chose PostgreSQL over MongoDB for the billing system because ACID transactions are non-negotiable for financial data"
• "Decided to use a monorepo after evaluating multi-repo coordination overhead"
• "Selected Kafka over RabbitMQ for event streaming due to replay capability"

Goal (0.80)

Goals are objectives and aspirations — things you are working toward. They decay somewhat faster than decisions because goals can shift as circumstances change.

Examples:

• "Migrate the authentication system to OAuth 2.0 by end of Q2"
• "Achieve 95% test coverage on the core payment module"
• "Learn Rust well enough to contribute to the compiler project"

Experience (0.75)

Experiences are notable events or processes worth remembering. They have moderate strength but faster decay — the specific details fade while the lessons persist.

Examples:

• "The production outage on March 5th was caused by a connection pool exhaustion under load"
• "The client demo went well — they were most impressed by the real-time collaboration feature"
• "Onboarding the new engineer took three weeks because documentation was outdated"

Learning (0.70)

Learnings represent new knowledge acquired during a session. They cover facts, techniques, and patterns that expand your understanding.

Examples:

• "TypeScript's satisfies operator preserves the narrowed type while checking assignability"
• "Next.js parallel routes allow independent loading states for different page segments"
• "The structuredClone() API handles circular references unlike JSON.parse(JSON.stringify())"

Relationship (0.70)

Relationships track connections between people, teams, projects, or systems. They share the same base strength as learnings but decay much more slowly because relationship context tends to remain relevant.

Examples:

• "Sarah from the platform team is the domain expert on the event pipeline"
• "The billing service depends on both the user service and the payment gateway"
• "The design system team maintains the component library and reviews all UI PRs"

Preference (0.60)

Preferences capture user style and personal choices. They start with lower strength but have very slow decay — your preferences rarely change rapidly.

Examples:

• "Prefer functional components with hooks over class components"
• "Use kebab-case for file names and camelCase for variables"
• "Prefer explicit error handling over try-catch-all patterns"

Observation (0.40)

Observations are casual facts or notices that may or may not prove useful. They start weak and decay quickly — if they turn out to be important, recalling them will strengthen them through spaced reinforcement.

Examples:

• "The CI pipeline takes about 12 minutes for a full run"
• "The staging database is on version 15.2 of PostgreSQL"
• "The team has a standup at 10am EST on weekdays"

How Types Are Assigned

When you run /brain:memorize, the agent analyzes the content and automatically assigns the most appropriate type based on the nature of the information:

• Decisions are identified by language like "chose", "decided", "went with", "selected"
• Insights involve patterns, realizations, or deep understanding
• Goals involve future-oriented language and objectives
• Experiences describe events, incidents, or processes
• Learnings introduce new facts, APIs, or techniques
• Relationships describe connections between entities
• Preferences express personal style or choices
• Observations are factual notes without strong significance

The agent can also assign types based on the context of the conversation, not just keyword matching. You do not need to specify the type manually.