In use now

Best forOne command, a model, and a chat box.

The catchOllama normally unloads idle models after about five minutes unless configured otherwise.

Best forThe runtime on the IronClaw machine. llama-server is what actually serves GPT-OSS.

The catchThe flags. There are a lot of flags, and two of them set the context.

Best forA document chat in front of a local model. It sat in the early stack, between Ollama and llama.cpp.

The catchAnother program between you and the weights. I did not keep it as the front door.

Best forThe browser I actually talk to Lucy through.

The catchIt is the front door. The model and AgentGate are still doing the work behind it.

Best forA desktop app when you do not want a flag in sight.

The catchI have not run it. The model browser is the thing people notice first, and downloads add up.

Best forA local chat that installs like a normal app.

The catchI have not run it. The model list tends to lag whatever just came out.

Best forAn open desktop client with a local server under it.

The catchI have not run it. Extensions are a second project. Start with the chat.

Best forA browser front end in front of Ollama.

The catchI have not run it. You would be running two things, and both of them need an update.

Best forImage graphs you can actually see.

The catchI have not run it. The first graph is a project of its own, and custom nodes will fight.

Best forSpeech to text on the machine you already have.

The catchI have not run it. Long audio is where a small model usually starts guessing.