Goose: an honest local agent

I wanted an agent - something I could hand a task to, that would read files, run commands, search the web, and get it done - but running on my models, with my work staying on my own machine. Goose, an open-source agent from Block, is what I settled on.

Let me be honest up front, because that is the whole point of this site: it is a reliable backup, not a marvel. It is not as smooth as the polished cloud agents. But it runs entirely on my hardware, on my models, and it gets real work done. For private, local, yours - it is more than good enough.

How it fits my stack

Goose does not talk to my models directly. It talks to my gateway, and the gateway does the rest:

  Goose  --->  LiteLLM (:4000)  --->  my local models (vLLM / Ollama)

I point Goose’s “OpenAI” provider at the gateway and pick the model I want it to think with - for me, my agentic flagship, a coding-focused model that is good at using tools. That is the entire connection. Because the gateway speaks the standard API, Goose thinks it is talking to OpenAI; it is really talking to my box.

Then I turn on a few of its built-in abilities: reading and writing files and running shell commands, read-only access to my code on GitHub, and web search. Each of those that needs a key keeps its key outside the project, never in anything I would share.

The one thing that makes or breaks it

An agent is only useful if the model can actually call tools - say “run this command” in a way the agent understands. If that translation is not set up, the model will look like it is working (“thinking…”), and then do nothing at all. I wrote about that silent stall in the two models I run; it is the single most common reason a local agent appears broken when the model is fine. Get the tool-calling right first, and Goose comes alive.

What it is genuinely good at

• Private by default. The agent that touches my files and runs my commands is powered by a model on my own machine. My code never leaves the room.
• Real tasks, not just chat. It edits files, runs things, checks results, and keeps a little to-do list as it goes.
• Free to run and always there. No per-task cost, no account, works offline.

Where it falls short (the honest part)

• It is slower and less polished than the big cloud agents. You feel the difference.
• The local model is less capable than a frontier cloud one, so it sometimes needs a clearer instruction or a second try.
• It occasionally fumbles a tool call and needs a nudge.

None of that is a dealbreaker for me. It is the difference between a car you own outright and a chauffeur you rent. The chauffeur is smoother. But the car is yours, it is always in the drive, and everywhere it takes you stays with you.

You do not need my exact models. Goose runs against a small Ollama model on a laptop just as happily:

• Point it at your gateway (or straight at Ollama) and pick a model that can call tools.
• Set the tool-calling up before you judge it - most “it does nothing” problems are that, not the model.
• Expect a reliable backup, not a miracle - and value it for what it is: an agent that is private, free, and entirely yours.