The NetworkTechnology
AI Integration
Stateful AI agents built on Letta, with three-layer knowledge and tier-based access.
The AI layer in ƒxyz is built around Fixies -- stateful agents running on the Letta framework (formerly MemGPT). Each agent maintains persistent memory across conversations and has access to a three-layer knowledge system:
- Letta Archives -- long-term document knowledge (brand, compliance, market data, graph schema)
- Graphiti temporal graph -- conversation-extracted entities and facts stored in FalkorDB
- Per-agent memory -- structured blocks (persona, human, context, preferences) that stay in-context
Fixies are scoped to three categories:
| Category | Scope | Examples |
|---|---|---|
| Network Fixies | System-wide, shared across all members | Florin (market), Cypher (graph queries), Sentinel (risk) |
| Member Fixies | Personal to each member | Auto-created on signup, up to 3 per member |
| Project Fixies | Circle-scoped | Tied to a specific circle's context |
Agent capability scales with member contribution tier: free (Haiku), standard (Sonnet), and premium (Opus). All tiers have access to the same 16 tools -- differentiation is in model quality and memory depth.
Detailed Reference
- The Fixies -- full architecture, surface routing, tool definitions, and tier system