ClawVault
Core Concepts

Memory Types

Eight structured memory types for organizing your agent's knowledge

ClawVault organizes memories into 8 structured types, each designed for specific kinds of information. This classification helps with retrieval, context injection, and maintaining clean mental models.

The 8 Memory Types

fact

Static information that doesn't change over time.

Examples:

  • API endpoints and their parameters
  • System configurations
  • Technical specifications
  • Contact information
  • Process documentation
clawvault remember fact "Stripe API Rate Limit" --content "100 requests per second per API key"

feeling

Emotional context, sentiment, and subjective experiences.

Examples:

  • How a meeting felt
  • Confidence levels in decisions
  • Team morale observations
  • User feedback sentiment
clawvault remember feeling "Client Demo Went Well" --content "Team felt confident, client asked good questions, positive energy"

decision

Choices made with context and reasoning.

Tip: Decisions automatically receive structural importance (>= 0.8) in observational memory.

Examples:

  • Architecture choices
  • Tool selections
  • Process changes
  • Strategic directions
clawvault remember decision "Use PostgreSQL over SQLite" --content "Need concurrent writes for multi-agent setup. SQLite locks cause bottlenecks."

lesson

Things learned from experience, mistakes, or insights.

Examples:

  • What worked/didn't work
  • Process improvements discovered
  • Debugging insights
  • Best practices learned
clawvault remember lesson "Always backup before schema changes" --content "Lost 2 hours of data during migration testing. Automated backups now required."

commitment

Promises, deadlines, and obligations.

Examples:

  • Project deadlines
  • Meeting promises
  • Feature deliverables
  • Personal commitments
clawvault remember commitment "Demo ready by Friday" --content "Promised client working prototype with user auth and basic dashboard"

preference

Subjective choices and preferred ways of working.

Examples:

  • Code style preferences
  • Tool preferences
  • Communication styles
  • Work patterns
clawvault remember preference "Async standups work better" --content "Team prefers written updates over video calls. More thoughtful, less interruption."

relationship

People, their roles, and how to work with them.

Examples:

  • Team members and their strengths
  • Client contacts and preferences
  • Stakeholder relationships
  • Communication patterns
clawvault remember relationship "Sarah Chen - Product Manager" --content "Detail-oriented, prefers data-driven decisions. Responds best to Slack, not email."

project

Active work, goals, and project context.

Examples:

  • Current initiatives
  • Project status updates
  • Goals and objectives
  • Work streams
clawvault remember project "User Dashboard Redesign" --content "Q1 initiative. Focus on mobile-first, reduce clicks by 30%, A/B testing framework ready."

Why Types Matter

Better Retrieval

# Find all decisions about databases
clawvault search "database" -c decisions

# See all active commitments
clawvault list commitments

Context-Aware Injection

Different memory types surface at different times:

  • Decisions → Critical for planning sessions
  • Commitments → Important for deadline tracking
  • Relationships → Key for meeting preparation

Graph Connections

Types create meaningful graph relationships:

# In people/sarah-chen.md
Related decisions: [[database-choice]], [[framework-selection]]
Active projects: [[dashboard-redesign]]

Quick Capture vs Structured Storage

For quick notes that you'll process later:

clawvault capture "TODO: Review PR tomorrow"

When you know the type:

clawvault remember decision "Use TypeScript" --content "Team voted, better DX"

Note: Quick captures go to the inbox/ category for later processing into proper types.

Auto-Classification

The observational memory system can automatically classify session content:

  • Decisions → detected from choice language ("decided to", "going with")
  • Lessons → detected from learning patterns ("learned that", "mistake was")
  • Relationships → detected from people mentions

This helps maintain structure even in conversational sessions.

On this page