agentchain

Why we included this project

AgentChain lets a large language model act as the planner for multi-step tasks, deciding which specialized agents or models to call and in what order. You describe a goal in natural language and the chain handles the handoffs, so you are not wiring a single monolithic prompt. It is fully multimodal, accepting text, image, audio, and tabular data as input and output, which fits pipelines that mix vision, speech, and language work. The codebase is compact and readable, a useful reference for how hierarchical tool composition can be structured, and adding custom agents for project-specific needs is straightforward.