Additional Tools Worth Knowing

Beyond the core stack, a few other tools come up regularly in local AI development. Two worth knowing about in the current ecosystem.

OpenCode

OpenCode is an AI coding assistant that runs in the terminal. Where tools like Kilocode integrate into a GUI editor, OpenCode is designed for developers who live in the terminal — you can interact with it directly from the command line, pipe output into it, and use it in scripting workflows.

It supports local models via Ollama and OpenAI-compatible APIs, which means you can point it at whatever model you're running locally. The interface is minimal by design: you describe what you want, it generates code or edits, and you review the result.

For tasks like writing shell scripts, generating boilerplate, or asking quick implementation questions without leaving the terminal, OpenCode fits naturally into a keyboard-driven workflow. It also works well for code review tasks: pipe in a diff or a file, ask it to identify issues, get structured feedback back.

One practical use case worth highlighting for the context of this series: once you have Ollama running with Tailscale, you can use OpenCode on a remote machine pointed at your local model server. Your workstation does the inference, your laptop does the interaction.

Oh My Open Agent

Oh My Open Agent is a framework for building and managing local AI agent configurations — think of it as a structured way to define what tools your agent has access to, what its default behaviour is, and how it interacts with your local environment.

The name is a deliberate nod to Oh My Zsh: the goal is to make agent configuration as composable and community-driven as shell configuration has become. You can pull in community-maintained agent profiles, customise them, and contribute your own.

It's particularly useful if you're running multiple agents for different purposes — one for code review, one for research, one for document processing — and want a clean way to manage their configurations without maintaining separate scripts or environment setups for each.

As of writing, it's still early-stage software, but the approach is sensible and it integrates well with the rest of the tooling covered in this series.