code-cli-tunnel

Prepare your remote Compute Instance with code-cli tunneling capability

Install micromamba

Recommended install method (official docs): https://mamba.readthedocs.io/en/latest/installation/micromamba-installation.html

Run this in your remote instance (bash):

"${SHELL}" <(curl -L https://micro.mamba.pm/install.sh)
# accept defaults (bin: ~/.local/bin, init shell: yes, root prefix: ~/micromamba, conda-forge: yes)
source ~/.bashrc
micromamba --version

If you skipped shell init (or need to re-run it):

~/.local/bin/micromamba shell init --shell bash --root-prefix ~/micromamba
source ~/.bashrc

Non-interactive bash (CI / scripts / bash -c)

Non-interactive bash does not source ~/.bashrc, so ~/.local/bin may be missing from PATH.

Run the env patch once to create a persistent symlink in /usr/local/bin:

./noninteractive-bash-env.sh
bash -c 'command -v micromamba && micromamba --version'

If you cannot create symlinks in /usr/local/bin, you can still use:

BASH_ENV="$(pwd)/noninteractive-bash-env.sh" bash -c 'micromamba --version'

Downgrade base to Python 3.12 and register a Jupyter kernel

micromamba install -n base -c conda-forge "python=3.12" ipykernel anywidget notebook plotly -y
micromamba run -n base python --version
micromamba run -n base python -m ipykernel install --user --name micromamba-base --display-name "Python (micromamba base)"
jupyter kernelspec list

Web Search in Codex

web_search is deprecated; use the web_search_request feature flag instead.

in the ~/.codex/config.toml

[features]
web_search_request = true        # allow the model to request web searches