The agent runtime environment. Each agent operates as an independent cognitive process with its own context window, decision-making logic, and specialized domain expertise.
Components:
Text is encoded into 2560-dimensional semantic vectors using Qwen3-4B, enabling semantic search, similarity matching, and context-aware retrieval across all knowledge domains.
Model:
Qdrant vector database stores all embeddings with metadata. Collections are domain-segmented, enabling targeted queries and namespace isolation across agents and knowledge types.
Engine:
PARA (Projects, Areas, Resources, Archive) is the organizational backbone of the Hive Mind. Every piece of knowledge, every task, every document maps to one of these four categories — creating a self-consistent taxonomy that scales.
Active work with deadlines. Specific outcomes, time-bound, and actively being pursued. The燃烧 of the system.
Ongoing responsibilities without a deadline. Long-term maintenance, recurring duties, and domain ownership.
Reference material and topical knowledge. Information you want to retain but aren't actively working on.
Completed or inactive items. Stuff that's no longer relevant but worth preserving for historical context.
Every collection in Qdrant represents a namespace of semantic knowledge. Each is domain-segmented for targeted retrieval. 50+ collections covering the full breadth of the Hive Mind's awareness.