AgentDock

Why we included this project

AgentDock takes a node-based approach to agent building: every capability is a node, and tools are specialized nodes, so you assemble workflows instead of wiring together scattered function calls. Its core idea is configurable determinism, meaning you decide which steps run on model inference and which follow fixed logic, a trade-off that matters when you want systems that behave predictably with minimal supervision. The project is TypeScript throughout, pairing a backend-first core with a Next.js client that shows how the framework fits into a real product, and it ships ready-made agents worth studying, including a multi-stage diagnostic reasoning agent and a seven-tool cognitive reasoning engine. For teams that want type safety and control over their own plumbing rather than a one-size-fits-all platform, this is a useful reference before you hand-roll your own agent infrastructure.