eino

Eino is a Go framework for building applications that use large language models (AI systems that process and generate text). It comes from ByteDance's CloudWeGo team and is designed specifically for Go developers who want to build AI-powered features without having to handle all the low-level coordination themselves.

The framework provides three main layers. The first is a set of reusable components: standard building blocks for connecting to chat models (OpenAI, Claude, Gemini, Ollama, and others), searching a knowledge base, generating text embeddings, and similar tasks. These components follow Go conventions and have consistent interfaces so they can be swapped or combined.

The second layer is a composition system for wiring components together into workflows. You define a graph of nodes, each responsible for a step, and the framework handles running them in order, passing data between steps, and managing streaming output throughout the pipeline. A workflow built this way can also be exposed as a tool that an AI agent can call.

The third layer is an Agent Development Kit for building autonomous agents, meaning programs where the AI decides what steps to take based on a goal rather than following a fixed sequence. A simpler agent handles single-model tasks with optional tools. A deeper agent can break a complex problem into pieces, delegate to specialized sub-agents, run shell commands, execute Python, and search the web.

Additional features include built-in support for pausing an agent mid-task to ask a human for input before continuing, callback hooks for logging and monitoring, and a visual debugging tool. The framework requires Go 1.18 or later and is available under an open-source license.