Chat works but not opencode/claude

[axys16]

Hi.

I have installed the last Lemonade. I have tested the Chat with several Model, all work very well.
(llama.cpp vulkan backend / Ubuntu 26.04)

But impossible to have a working opencode or claude.

I run
lemonade run Qwen3-Coder-30B-A3B-Instruct-GGUF
then
lemonade launch opencode -m Qwen3-Coder-30B-A3B-Instruct-GGUF

The prompt do nothing, 0 token used. I see no error on opencode logs.

Do I miss something ?

[VectorPeak]

If the built-in Chat works with that same model/backend, the llama.cpp/Vulkan side is probably fine. The 0 token used symptom usually points one layer above that: the agent is not actually sending a completion request to Lemonade, or it is using a stale/wrong provider config. What I would check first is launching with the explicit long-form model flag, just to remove any CLI parsing ambiguity:

lemonade load Qwen3-Coder-30B-A3B-Instruct-GGUF

lemonade launch opencode --model Qwen3-Coder-30B-A3B-Instruct-GGUF
lemonade launch claude --model Qwen3-Coder-30B-A3B-Instruct-GGUF

For opencode, Lemonade writes/updates the config under:

~/.config/opencode/opencode.json

Please check that this file contains a Lemonade provider pointing at your local server, usually http://127.0.0.1:13305/v1, and that the selected model is exactly Qwen3-Coder-30B-A3B-Instruct-GGUF. If you had used opencode before, a stale config is a very plausible cause. A quick test is to move it aside and let Lemonade regenerate it:

mv ~/.config/opencode/opencode.json ~/.config/opencode/opencode.json.bak
lemonade launch opencode --model Qwen3-Coder-30B-A3B-Instruct-GGUF

Also worth checking whether an API key mismatch is involved. If you have LEMONADE_API_KEY set, the launched agent must use the same key. If it fails, the error from that curl should be more useful than the agent UI. For Claude Code specifically, make sure you are launching it through Lemonade and not starting claude separately in a shell that still has Anthropic/OpenAI-related env vars pointing somewhere else. The intended path is:

lemonade launch claude --model Qwen3-Coder-30B-A3B-Instruct-GGUF

Could you paste these outputs, with any keys redacted? ```bash
lemonade --version
opencode --version
claude --version
lemonade status
lemonade list --downloaded

And for `opencode`, the generated provider/model section from:

```bash
cat ~/.config/opencode/opencode.json

The important clue will be whether Lemonade logs show an incoming /v1/chat/completions or /v1/messages request when you press enter in the agent. If there is no request at all, it is almost certainly provider/config/env. If there is a request and it returns an error, the server log line should tell us the next thing to fix.

---

If my answer solved your problem, you can click answered the question. I'm really here to help, and along the way I'm also collecting Galaxy Brain badges haha 😆