I saw a colleague trying to bring in and use Claude Code at work. They struggled to connect Claude Code with a pre-configured LiteLLM endpoint in an air-gapped environment, and it ended up failing. In the end, they are currently using @goose_oss lol. He realized just how convenient it is. Goose offers compatibility with a truly wide variety of model providers.

The @ollama cloud subscription gives broad access to open-weight models, and usage vs. credits is fairly generous. However, if you run kimi-k3 or glm-5.3 as the main model, the allocation burns through fast. @goose_oss is the harness that fits this plan best. Keep the main agent on deepseek-v4.1-flash and define subagents as goose recipes. A recipe is the subagent config. Lock role, extensions, and model into the recipe, and the subagent spins up with that exact setup when invoked. Role split: • Hard, long-horizon reasoning / difficult coding: glm-5.3 • Document and external-repo analysis: glm-5.3-flash • Web search: a cheap recipe with search tools attached (deepseek-v4.1-flash or glm-5.3-flash) Put into .goosehints when to call which recipe and when the main agent should handle the work itself. As the task proceeds, the main agent only invokes the recipes it needs. Expensive models get used only for those segments, so you save tokens. Other harnesses often support per-subagent models too, but wiring Ollama Cloud’s open-weight models into reusable recipes is where goose fits best.

Most people live with agents — inside CLI, TUI, and Desktop UI. We need to learn to operate agents from outside the agent. Go headless, dive into the SDK, and wire it into CI/CD. Assign different LLMs based on the agent's role. Orchestrate sub-agents. @goose_oss can do all of this. Check it out: And look at what other harness users are building too. That's totally fine — they've documented so many use cases and it's super helpful. Using a great agent is good. But to develop an eye for what makes an agent actually good, you have to build one yourself. Only someone who has baked cookies knows how much sugar and butter goes in. Someone who has only eaten them never will.

To be completely honest, it’s the best harness I’ve ever experienced, so I’ve just been learning from it and can’t really offer any feedback. In particular, the models in Droid Core were so impressive—they often exhibit behaviors I’ve never experienced in other harnesses, to the point where I wonder if they are truly the models I thought I knew. And Agent Readiness makes me want to subscribe just for that feature alone...

It is important for an agent to know itself. Only then can it make full use of its underlying harness. @goose_oss uses the goose-doc-guide skill to reference what it is capable of. The problem, however, is that in air-gapped network environments like mine, it cannot access the outside. So I raised this issue with the goose team, and @alexjhancock took care of implementing a fix. I want to thank the goose team for developing responsibly as a standard agent of the @AgenticAIFdn . Thanks to them, goose now knows who it is and does its job well, even in air-gapped networks.

The @goose_oss CLI offers a wide range of commands, and one of them is `goose review`. By default, it reviews the diff between your working tree and HEAD, but you can also specify a range like `main...HEAD`. Most agents provide code review through a slash command like `/review` inside their CLI or TUI. That's how I've been using it too. What sets goose apart is that this review feature can be invoked headlessly as an external process. That makes it easy to drop directly into a CI/CD pipeline. An example is `CodeGoose`, a sample project I built. It runs `goose review` inside a GitHub Actions workflow to review PR diffs and leave comments automatically. It works quite reliably, and since goose supports BYOK (Bring Your Own Key), you can build your own fully self-owned review agent. As GDK (Goose Development Kit), which is currently under active development, matures, I think it will be possible to build a much more lightweight yet specialized review agent.

As I’ve mentioned before, work in a company follows established workflows. Delegating tasks to an agent in a company means that the agent must work as a member of that organization. @goose_oss recipes allow us to define such internal workflows and carry out a consistent work process every time. It carries out tasks with astonishing consistency each time. Through recipes, we can delegate our work to agents with much greater confidence.