This docker container has preinstalled versions of several LLM cli-tools. Running in a container can be a convenient way to control the limit the scope of what the LLM can access on your machine without having to trust the developers.
Below we also document how to configure these to access various models, including free models and open source models.
The following interfaces are preinstalled:
- claude code: https://github.com/anthropics/claude-code
- copilot cli: https://github.com/github/copilot-cli
- codex-cli: https://github.com/openai/codex
- cline-cli: https://github.com/cline/cline/tree/main/cli
Because copilot is free with a GitHub account (and 'premium' models are free to .edu accounts), it is a good starting point:
docker run --rm -ti ghcr.io/schmidtdse/llm copilot
Copilot can also be configured to work with various local / open models, including the models hosted by NRP
Add configuration here by volume linking a `.copilot/config.toml` file
Uses the opencode.json config provided in this repo to access NRP models.
docker run --rm -ti -e NRP_API_KEY=$NRP_API_KEY ghcr.io/schmidtdse/llm opencode
Cline is an open-source vscode extension that also has a command-line tool. Though development has focused on Anthropic's models, cline is the only one on this list that is unaffiliated with any specific AI lab and has excellent support for a broad base of models.
docker run --rm -ti -e ANTHROPIC_BASE_URL=https://ellm.nrp-nautilus.io -e ANTHROPIC_MODEL='glm-4.6' -e ANTHROPIC_AUTH_TOKEN=$ANTHROPIC_AUTH_TOKEN ghcr.io/schmidtdse/llm claude