Penfield

Penfield Memory

Persistent memory that compounds. Your agent remembers conversations, learns preferences, connects ideas, and picks up exactly where it left off—across sessions, days, and channels.

Tools

Memory

Knowledge Graph

Context & Analysis

Artifacts

Writing Memories That Actually Work

Memory content quality determines whether Penfield is useful or useless. The difference is specificity and context.

Bad — vague, no context, unfindable later:

"User likes Python"

Good — specific, contextual, findable:

"[Preferences] User prefers Python over JavaScript for backend work.
Reason: frustrated by JS callback patterns and lack of type safety.
Values type hints and explicit error handling. Uses FastAPI for APIs."

What makes a memory findable:

1. Context prefix in brackets: [Preferences], [Project: API Redesign], [Investigation: Payment Bug], [Decision]
2. The "why" behind the "what" — rationale matters more than the fact itself
3. Specific details — names, numbers, dates, versions, not vague summaries
4. References to related memories — "This builds on [earlier finding about X]" or "Contradicts previous assumption that Y"

Memory Types

Use the correct type. The system uses these for filtering and analysis.

Importance Scores

Use the full range. Not everything is 0.5.

Connecting Memories

Connections are what make Penfield powerful. An isolated memory is just a note. A connected memory is understanding.

After storing a memory, always ask: What does this relate to? Then connect it.

Relationship Types (24)

Knowledge Evolution: supersedes · updates · evolution_of Use when understanding changes. "We thought X, now we know Y."

Evidence: supports · contradicts · disputes Use when new information validates or challenges existing beliefs.

Hierarchy: parent_of · child_of · sibling_of · composed_of · part_of Use for structural relationships. Topics containing subtopics, systems containing components.

Causation: causes · influenced_by · prerequisite_for Use for cause-and-effect chains and dependencies.

Implementation: implements · documents · tests · example_of Use when something demonstrates, describes, or validates something else.

Conversation: responds_to · references · inspired_by Use for attribution and dialogue threads.

Sequence: follows · precedes Use for ordered steps in a process or timeline.

Dependencies: depends_on Use when one thing requires another.

Recall Strategy

Good queries find things. Bad queries return noise.

Tune search weights for your query type:

Filter aggressively:

• memory_types: ["correction", "insight"] to find discoveries and corrections
• importance_threshold: 0.7 to skip noise
• enable_graph_expansion: true to follow connections (default, usually leave on)

Workflows

User shares a preference

penfield_store({
  content: "[Preferences] User wants responses under 3 paragraphs unless complexity demands more. Dislikes bullet points in casual conversation.",
  memory_type: "fact",
  importance: 0.8,
  tags: ["preferences", "communication"]
})

Investigation tracking

// Start
penfield_store({
  content: "[Investigation: Deployment Failures] Reports of 500 errors after every Friday deploy. Checking release pipeline, config drift, and traffic patterns.",
  memory_type: "task",
  importance: 0.7,
  tags: ["investigation", "deployment"]
})

// Discovery — connect to the investigation
discovery = penfield_store({
  content: "[Investigation: Deployment Failures] INSIGHT: Friday deploys coincide with weekly batch job at 17:00 UTC. Both compete for DB connection pool. Not a deploy issue — it's resource contention.",
  memory_type: "insight",
  importance: 0.9,
  tags: ["investigation", "deployment", "root-cause"]
})
penfield_connect({
  from_memory_id: discovery.id,
  to_memory_id: initial_report.id,
  relationship_type: "responds_to"
})

// Correction — supersede wrong assumption
correction = penfield_store({
  content: "[Investigation: Deployment Failures] CORRECTION: Not a CI/CD problem. Friday batch job + deploy = connection pool exhaustion. Fix: stagger batch job to 03:00 UTC.",
  memory_type: "correction",
  importance: 0.9,
  tags: ["investigation", "deployment", "correction"]
})
penfield_connect({
  from_memory_id: correction.id,
  to_memory_id: initial_report.id,
  relationship_type: "supersedes"
})

Session handoff

penfield_save_context({
  memory_ids: [discovery.id, correction.id, initial_report.id],
  session_id: "deployment-investigation-2026-02"
})

Next session or different agent:

penfield_restore_context({
  checkpoint_id: "checkpoint-uuid",
  merge_mode: "append"
})

What NOT to Store

• Verbatim conversation transcripts (too verbose, low signal)
• Easily googled facts (use web search instead)
• Ephemeral task state (use working memory)
• Anything the user hasn't consented to store about themselves
• Every minor exchange (be selective — quality over quantity)

Tags

Keep them short, consistent, lowercase. 2–5 per memory.

Good: preferences, architecture, investigation, correction, project-name Bad: 2026-02-02, important-memory-about-deployment, UserPreferencesForCommunicationStyle

Also Available Outside OpenClaw

The native OpenClaw plugin is the fastest path, but Penfield works with any AI tool anywhere:

Claude Connectors

Name: Penfield
Remote MCP server URL: https://mcp.penfield.app

Claude Code

Claude mcp add --transport http --scope user penfield https://mcp.penfield.app

MCP Server — for Gemini CLI, Cursor, Windsurf, Intent, Perplexity Desktop or any MCP-compatible tool:

{
  "mcpServers": {
    "penfield": {
      "command": "npx",
      "args": [
        "mcp-remote@latest",
        "https://mcp.penfield.app/"
      ]
    }
  }
}

API — direct HTTP access at api.penfield.app for custom integrations.

Same memory, same knowledge graph, same account. The plugin is 4-5x faster (no MCP proxy layer), but everything stays in sync regardless of how you connect.

Links

• Plugin: openclaw-penfield on npm
• Source: github.com/penfieldlabs/openclaw-penfield
• Sign up: portal.penfield.app/sign-up
• Website: penfield.app
• X: @penfieldlabs