Changelog#

0.2.0 (2026-06-22)#

Added#

  • Default data directory for new installs moved from ~/.cg/ to ~/.conda/global/. Existing installs continue using ~/.cg/ until explicitly migrated. (#9)

  • Manifest moved from inside the data directory to ~/.conda/global.toml, a flat file alongside the data directory for easy discoverability.

  • New conda global migrate subcommand performs a reinstall-based migration from ~/.cg/ to the new layout. Copies the manifest, runs a fresh sync, then removes the old directory.

  • conda global uninstall <env> is now the primary uninstall form. The previous conda global uninstall -e <env> form remains supported, and conflicting positional/flag targets are rejected. (#21)

  • Package-owned executables are exposed on install when conda package metadata lists them, so packages such as httpie expose http, httpie, and https by default. (#22)

Changed#

  • Install and environment creation now use the channels from conda configuration when no channel is provided on the command line or in the manifest. Explicit -c/--channel values are searched before configured channels, and --override-channels keeps the solve limited to explicitly supplied channels. (#20)

  • New installs persist the resolved channel list in global.toml, matching the solver inputs used for the initial install. (#20)

  • Documentation now compares conda-global with conda-exec and clarifies the new default data paths. (#14)

Removed#

  • Removed conda global run. Use conda exec for one-off execution from temporary or cached conda environments.

Security#

  • Environment and binary names are validated before filesystem access to reject path traversal and invalid tool names. (#10)

Packaging and CI#

  • Bump conda-trampoline, the split conda recipe, and release metadata to 0.2.0.

  • Test every supported Python version, 3.10 through 3.14, across Ubuntu, macOS, and Windows. (#23)

  • Update GitHub Actions dependencies, including Codecov v7. (#13, #15)

0.1.1 (2026-04-01)#

Changes#

  • Move repository to conda-incubator organization (governance#370).

  • Update all project URLs to conda-incubator/conda-global.

  • Simplify installation docs: remove pixi/mamba alternatives, mention standalone cg alias.

  • Bump conda-trampoline wheel/sdist version to 0.1.1 (it was still 0.1.0 in Cargo.toml / pyproject.toml, so wheels did not match the tag).

0.1.0 (2026-04-01)#

Initial release of conda-global and conda-trampoline.

Features#

  • Global tool management: install CLI tools into isolated conda environments and expose them on PATH, similar to pipx or pixi global but for the full conda ecosystem.

  • Rust trampoline launcher (conda-trampoline): a small binary that reads a JSON config and launches the real tool with zero conda activation overhead.

  • Manifest-driven: all state is tracked in a single TOML file, making it easy to share, back up, and version-control your global tool set.

  • conda plugin: registers as conda global via pluggy, so all commands are available as conda subcommands.

  • Standalone CLI: also available as the cg command for convenience.

Commands#

  • conda global install — install a tool into an isolated environment

  • conda global uninstall — remove a tool and its environment

  • conda global add / remove — manage dependencies in an existing tool env

  • conda global list — list installed tools

  • conda global update — update one or all tools

  • conda global sync — reconcile filesystem state with the manifest

  • conda global expose / hide — control which binaries are on PATH

  • conda global run — run a tool without installing it

  • conda global tree — show dependency tree for a tool env

  • conda global pin / unpin — lock or unlock a tool from upgrades

  • conda global ensurepath — add the bin directory to your shell PATH

  • conda global edit — open manifest.toml in your editor

  • conda global status — show summary of installed tools and PATH status

Packages#

This project ships as two packages:

  • conda-global — the Python plugin and CLI

  • conda-trampoline — the Rust trampoline binary, distributed as platform-specific wheels via maturin

Platforms#

  • Linux (x86_64, aarch64)

  • macOS (x86_64, Apple Silicon)

  • Windows (x86_64)

On this page
Show Source