Skip to content
tulpa docs

Dunbar Layers

The concept

tulpa organizes your network into layers inspired by Dunbar’s number research — the idea that humans maintain relationships at predictable scales. Each layer reflects a different depth of relationship:

  • Inner circle (≈5): Your closest collaborators and confidants. High-context, high-trust.
  • Active (≈15): People you work with regularly and keep in frequent contact with.
  • Sympathy (≈50): People you’d make time for — strong professional relationships you maintain deliberately.
  • Recognition (≈150): People you know well enough to have a real conversation with but don’t actively maintain.
  • Known (≈500+): People you’ve met or connected with but rarely interact with.

These numbers are approximate. Your actual layer sizes will vary based on how you use tulpa and the density of your network.

How layers affect agent behavior

Your agent uses Dunbar layers to make decisions about priority, autonomy and resource allocation:

  • Inner circle and active contacts get faster response times, richer coordination and more proactive outreach from your agent.
  • Sympathy-layer contacts receive steady but less frequent attention — your agent monitors relationship health and suggests touchpoints.
  • Recognition-layer contacts are maintained passively. Your agent tracks major changes (new roles, signals) but doesn’t proactively engage.
  • Known contacts are indexed but mostly dormant. Your agent activates these relationships only when a specific need arises — like discovering a warm path.

Layer assignment isn’t permanent. As interaction patterns change, contacts naturally drift between layers. Your agent recalculates positions based on communication frequency, meeting patterns and coordination activity.

How extensions see layers

Extensions (third-party integrations) don’t get access to your entire network. They see a filtered view based on Dunbar layers and your permission settings. You control which layers an extension can access, ensuring that casual connections aren’t exposed to tools that only need your active collaborators.

Connection to relationship health

Dunbar layers and relationship health work together. A contact in your sympathy layer whose relationship health is declining might trigger a suggestion from your agent — a check-in, a shared article or a meeting. Layers set the baseline expectations; health scores track whether you’re meeting them.