Local model runs | Versalist Documentation

The Versalist CLI sends public challenge cases to Ollama on your computer. Versalist judges the returned answers and records the Episode trace.

Requirements

Node.js 18.17 or later
Ollama installed and running on your computer
One installed Ollama model, such as Gemma, GPT-OSS, or DeepSeek
A Versalist API key
A challenge that allows local execution

Create an API key in API key settings. Set the key in your shell before you run a challenge.

export VERSALIST_API_KEY=vk_live_...

Run a local model

Configure the local Ollama endpoint.

versalist local configure ollama --url http://127.0.0.1:11434

List the models installed in Ollama.

versalist local models

Check the local model and Versalist connection.

versalist local doctor --model ollama:gemma3:4b

Run a challenge with the selected model and skill bundle.

versalist challenge run <challenge-slug> \
  --skill <skill-bundle-uuid> \
  --model ollama:gemma3:4b

Model format

Use the installed Ollama model tag after the ollama: prefix.

--model ollama:<installed-model-tag>

The CLI sends the selected model tag with the run. The platform does not verify the model identity. The trace labels the run as local execution and self-reported model provenance.

Run limits

The challenge must allow local execution.
The CLI sends public cases only.
The platform keeps local runs private.
Local runs do not enter the trusted leaderboard.
The platform uses judge quota before local inference starts.

Troubleshooting

Ollama is not available

Start Ollama, then run the doctor command again.

versalist local doctor --model ollama:gemma3:4b

The model is not available

Check the model tag in Ollama, then use the exact tag in the run command.

versalist local models

The challenge does not allow local execution

Ask the challenge owner to enable local execution. The CLI cannot override this policy.

Read the CLI reference for other CLI commands and the API key guide for key scopes.