Dive

When should you pick Dive? When you're shipping a Go backend and need an agent orchestrator that lives inside your codebase, not on a remote server. It's Apache-licensed, so no vendor lock-in, and you can shape exactly how agents run. The framework's core abstraction — composing steps, calling LLMs, managing state — is minimal by design. That's a strength for teams that already have strong Go patterns and don't want a framework fighting them. When should you pass? If your team isn't Go-savvy, or you're prototyping and want a managed dashboard instead of wiring up an agent loop yourself. Dive gives zero UI, no hosted runtime, and no pre-built connectors to Slack, Notion, or similar. You'll be writing Go code for every integration. Compared to alternatives like LangChain (Python, heavy on abstractions) or Vercel AI SDK (TypeScript, opinionated), Dive sits in a narrower lane: pure Go, production-first, embeddable. The project is smaller and younger, but the team behind it (Deep Noodle) ships real client work with it, so it's not just a side project. In practice, Dive shines for internal automation — the kind of agents that process documents, run CLI scripts, or manage workflows inside a Go service. The company's own client projects (customer support agent, codebase transformation) are good examples of the sweet spot. One caveat: the documentation and community are still maturing. If you need extensive tutorials, plugin ecosystems, or quickstart templates, you'll find fewer resources than with larger frameworks. But for Go teams that know what they're doing, Dive is refreshingly direct.