free models on openrouter.ai appear to be pretty much unusable... that was a waste of time.
... and i think my ollama is borked due to nvidia container stuff not doing its thing?!? fixing nvidia drivers is pretty far down the list of things i want to do :\
good news :D nvidia was fine, wrong docker compose config file was running for some reason ¯\_(ツ)_/¯
gave qwen3:8b a run using opencode.ai but it seems to just crash and freak out on any simple command (eg /init)
left it to download 30b model overnight... pretty sure that will have a bad time/fail on my old gpu though.
maybe there is some other issue with my ollama server running tools (^～^;)ゞ
okay... so it's not actually possible to use opencode.ai with ollama?!?
tool calling appears to be borked?
- https://github.com/sst/opencode/issues/1034
- https://github.com/sst/opencode/issues/3122
waste of time ++
as per https://github.com/sst/opencode/issues/2362#issuecomment-3247382727
i've set the env var ollama server:
OLLAMA_CONTEXT_LENGTH=12288
and https://github.com/sst/opencode/issues/2362#issuecomment-3243343525
ive set the model config in opencode.json to:
"qwen3-coder:30b": {
"name": "qwen3-coder:30b",
"tools": true,
"tool_call": true
}
the robot has now successfully created an empty file... server currently sounds like a 747 trying to run `/init` and we about 15mins in.
quoll (√)