Agent failed to start: wcore exited with code 127 during init

[BaddBwoy]

I'm not putting this in as a bug because I sincerely believe it's user error.

I've installed wayland on a DGX Spark, and trying to run models through Ollama. They seem to be installed properly.

I have Claude working fine when I start a chat, but both Qwen 3.5 and Llama 3.1 9B give the error "Agent failed to start: wcore exited with code 127 during init"

Any ideas?

[VectorPeak]

This looks much more like an environment/init failure than a model-quality or prompt issue. On Linux, exit code 127 almost always means “command not found” or “the interpreter/loader for the command was not found”. Since Claude starts fine but the Ollama-backed models fail during wcore init, the first thing I would check is whether the process that launches wcore can actually see Ollama and the exact model tags you configured.

A quick sanity pass from the same shell/user that starts wayland would be:

which ollama
ollama --version
ollama list
ollama run <exact-model-name-from-wayland-config>
curl http://127.0.0.1:11434/api/tags

The “exact model name” part matters here. Ollama is pretty literal about tags. For example, if ollama list shows llama3.1:8b but the config says Llama 3.1 9B, llama-3.1-9b, or another display name, the UI/config may look right while the backend cannot resolve it. Same idea for Qwen - use the exact tag reported by ollama list.

The other common gotcha is that wayland may not be running in the same environment as your terminal. If it is launched via systemd, Docker, a desktop launcher, or another supervisor, it may have a different PATH, so ollama works interactively but is invisible to wcore. In that case, either make sure ollama is on the service/container PATH, or configure the provider to talk to the Ollama HTTP server directly instead of relying on the CLI.

If Docker or any container boundary is involved, also check the host address. 127.0.0.1:11434 from inside a container points at the container itself, not necessarily the host running Ollama. Depending on how you start it, you may need host networking or an explicit host mapping.

# from wherever wayland/wcore runs
curl http://127.0.0.1:11434/api/tags

If that fails from the wayland runtime but succeeds from your normal shell, that is probably the issue.

One separate possibility: agent mode may require capabilities that the selected local model/tag does not support well, such as tool calling or structured output. That usually shows up as a model/provider error rather than wcore exited with code 127, though, so I would not chase that until the basic init checks above pass.

The most useful details to narrow this down would be:

OS/arch on the DGX Spark:
wayland version/commit:
How wayland is launched: terminal, systemd, Docker, desktop app, etc.
Output of: ollama list
The exact model strings configured for Qwen and Llama:
The stderr/log lines immediately before "wcore exited with code 127 during init":

My bet is either an Ollama binary/path visibility issue or a mismatch between the configured model name and the actual Ollama tag. The fact that Claude works is a good clue: the app can start an agent in general, but the local Ollama provider path is failing very early.

---

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 😆