Overview
Smollama is a distributed AI agent system designed for resource-constrained devices like Raspberry Pi. It enables nodes to sense their environment through GPIO sensors and system metrics, remember observations with semantic search, think using local LLMs via Ollama, communicate with other nodes over MQTT, and sync data using conflict-free replicated data types (CRDTs) for seamless offline-first operation.
Key Features
Sense
Read GPIO sensors, system metrics, and environmental data from connected hardware.
Remember
Store observations and memories in SQLite with vector embeddings for semantic search.
Think
Process events using local LLMs via Ollama with a tool-based reasoning loop.
Communicate
Share state with other nodes via MQTT for distributed coordination.
Sync
Operate offline and merge data deterministically using append-only CRDT logs with Lamport timestamps.
Dashboard
FastAPI web interface with HTMX for live monitoring and interaction.