gokhantos/opencrow

Employs a supervisor process architecture where agents, scrapers, and subsystems run as isolated child processes with crash recovery and exponential backoff, enabling horizontal scaling and fault isolation. Leverages the Claude Agent SDK with an in-process MCP bridge for tool access (no network latency), session resumption for conversational continuity, and vector memory injection for cross-session context. Includes 31 specialized sub-agents for orchestrated workflows (architect → planner → backend/frontend → reviewer → security) and time-series market data indexed in QuestDB alongside conversational memory in Qdrant.

Available on npm.