Overview
Single-agent loops degrade on multi-part questions: context fills with intermediate work, and the model loses the thread of what it was originally asked. This system borrows the Plan → Reason → Act structure from Vision-Language-Action architectures and applies it to research, so decomposition and execution are separate concerns handled by separate agents.
Architecture
- Orchestrator decomposes the incoming query and routes each part to a specialist rather than answering it directly.
- Research and Reasoning specialists handle retrieval-shaped and inference-shaped subtasks respectively, then hand results back for synthesis.
- Dual-layer memory: ChromaDB provides semantic long-term recall; a sliding working-memory window with pruning keeps the active context bounded.
- Tool registry exposing Calculator, ArXiv and Search through OpenAI-compatible function calling, so tools are declared once and callable by any agent.
- Fully async on asyncio, so agent operations run concurrently instead of serializing behind each other.
Interface
It runs as an interactive CLI with explicit verbs: search for research, reason for analysis, stats to inspect memory and system state. Making the agent’s mode an explicit command rather than an inferred intent keeps the system debuggable: when a result is wrong, you know which path produced it.