AI Coding Agents on KLC#
KLC provides a container module for running command-line AI coding agents — Claude Code , OpenAI Codex CLI , and GitHub Copilot CLI — against your project files. The container isolates the agent’s runtime from the shared KLC environment and mounts only the project and data directories you specify, so the agent cannot read outside of them.
Warning
Coding agents send file contents and prompts to Anthropic’s, OpenAI’s, or GitHub’s servers. Follow your IRB protocol and Northwestern data governance policies before pointing an agent at a directory that contains identifiable human-subjects data or other restricted data. See Data Governance and IRB.
1. Install the CLI Tool#
Each CLI installs with a single curl command into ~/.local/bin — no environment setup required. Install whichever agent(s) you want to use.
See the GitHub Copilot CLI install instructions for macOS and Linux, or run:
curl -fsSL https://gh.io/copilot-install | bash
~/.local/bin/copilot --version
You need a GitHub account with access to Copilot.
See the Claude Code install instructions , or run:
curl -fsSL https://claude.ai/install.sh | bash
~/.local/bin/claude --version
You need a Claude account.
See the Codex CLI documentation , or run:
curl -fsSL https://chatgpt.com/codex/install.sh | sh
~/.local/bin/codex --version
You need a ChatGPT account with access to Codex.
If the version check above fails with “command not found,” add ~/.local/bin to your PATH:
export PATH="$HOME/.local/bin:$PATH"
2. Load and Launch the Container#
Load the container module and launch it with the agent you installed, passing the project directory and (optionally) a data directory to mount:
module load ai-agent-container
ai_agent_container -a claude /path/to/project /path/to/data:ro
# or
ai_agent_container -a codex /path/to/project /path/to/data:ro
# or
ai_agent_container -a copilot /path/to/project /path/to/data:ro
Replace /path/to/project and /path/to/data with the directories the agent needs access to — for example your project repository under /kellogg/proj/<your-netid>/ and a dataset path under /kellogg/data/. The agent cannot see anything outside the paths you pass in. The :ro suffix mounts the data directory read-only; default to it for reference data or licensed datasets so the agent can read them but never modify or delete them. Drop the suffix only if the agent needs to write to that directory.
3. Log in to Activate the Agent#
The first time you launch, you’re prompted to log in to your Anthropic, OpenAI, or GitHub account, depending on the agent. If no prompt appears, run the login command inside the agent’s session:
/login
Your login persists for subsequent sessions in the same environment.