Letta Platform
This agent runs on Letta — an open-source framework for building stateful LLM agents with persistent memory, tool use, and structured reasoning loops.
Agent Engine
Letta manages the agent lifecycle: system prompt, memory blocks, tool dispatch, and conversation history with automatic compaction.
Memory Architecture
Core memory blocks (in-context) + archival memory (persistent). Blocks have character budgets and are always in the prompt window.
Tool Surfaces
Agents invoke tools via structured calls. PALEO extends this with species-specific tools like simulate_instinct_decision and query_pot_wiki.
Recall & Compaction
Old messages are evicted and replaced with recursive summaries. All messages persist and remain retrievable via conversation search.
PALEO Agent Architecture
Live view of the Letta-based Instinct Agent internals — identity, tools, system prompt, memory blocks, and archival memory. Click a section in the left rail to explore.