Troubleshooting

Model server

The connection indicator is red

The indicator beside Models in the bottom left corner turns red when Artemis Desktop loses the connection to the model server.

1. Check that the model server is still running and look for errors in its terminal. For llama.cpp, also check the file set with --log-file.
2. Open Models, click Edit on the source and check the Address. It must match the host and port the server was started on, for example http://127.0.0.1:8081/.
3. If the server was started with an API key, enter the same key in API key.

The model server must keep running while you use Artemis Desktop. Closing its terminal window stops it.

The address is not accepted

Artemis Desktop checks the address as you type. If it is not accepted:

• Check that the model server is running.
• Check that the protocol is http:// for a local server unless you configured TLS.
• Check that the host and port match the ones the server was started on.

The port is already in use

If the model server fails to start because the port is in use, another application is listening on it. Start the server on a different port (--port for llama-server), then update the Address in Artemis Desktop to match.

The model server runs out of memory

If the model fails to load or the server crashes, the model and its context may not fit in your GPU or system memory. For llama.cpp:

• Reduce the context length set with -c. The examples use 90,000 tokens.
• Reduce the number of layers offloaded to the GPU with -ngl, so more of the model runs on the CPU.
• Use a smaller quantisation of the model, or a smaller model.

A shorter context limits how much code an agent can consider at once, and running more layers on the CPU is slower.

Responses are slow

Local inference speed depends on your hardware. Check that the model runs on the GPU (for llama.cpp, that layers are offloaded with -ngl) and that no other heavy workload, such as a Runner executing benchmarks on the same machine, is competing for the GPU or memory.

The model does not appear in Artemis

When the source is connected, its models appear in Artemis as <Name>/<alias>, for example Local/Qwen3-Coder-30B-A3B. For llama.cpp, the alias comes from the --alias option of llama-server. If no model is listed, check that the server is running and that the source shows Serving 1 model or more.

Runner and runs

The Runner is offline

Discover and Branch validation need an online Runner. Check that the Runner is started, that it is selected for the Project under Project Settings → Runner and Scripts, and see Runner troubleshooting for connection and proxy problems.

A Script fails on the Runner

Click Run on the Script under Project Settings → Runner and Scripts and read the logs of the failing Command. Common causes:

• The Runner's machine is missing a tool the Command needs, such as Python, CMake or a compiler.
• The Command uses python3, but on Windows the interpreter is usually python.
• A Command relies on a previous Command's cd or environment. Each Command runs in its own shell from the root of the code. See Runner and Scripts.

From the CLI, artemis changeset validation logs and artemis discovery versions logs show the same logs. See Artemis CLI.

A Discovery Run is slow or measures nothing

With a local model, each Version takes longer than with a hosted model. Start with a small Number of versions, such as 5, and avoid running other heavy workloads on the same machine.

Individual Versions can fail, which is normal. If a run finishes without measuring any Version, check the Runner, the Script and the model selection first.

A run uses a hosted model

Check the models chosen for the run. In the app, the Orchestrator model and the LLM-council are set in the Objective & settings step of Discover setup, and can be changed later on the run's Setup page. Choose your local model, for example Local/Qwen3-Coder-30B-A3B, for both. When a coding agent starts the run, ask it to use the local model for every model choice, including the review models.

CLI, agents and account

The CLI cannot connect

Run artemis status. It reports whether the CLI is configured, whether your API key is valid and whether each service is reachable. To change the deployment or key, run artemis login again. See Connect and authenticate.

The coding agent does not use the Artemis skills

Check that the skills are installed and loaded. In Claude Code, run /plugin list and look for the Artemis plugin, then run /reload-plugins or restart Claude Code. For Codex, Cursor and GitHub Copilot, rerun the npx skills add command and restart the agent. See Artemis Skills.

You cannot sign in

See Troubleshooting account issues.

Getting help

For other problems, write to support@turintech.ai.