envloop-cli

Independent Python 3.11+ project providing the existing el CLI, version 0.1.1. It builds and installs without the envloop SDK checkout or any third-party runtime dependencies. The distribution is envloop-cli; its private implementation package is el_cli, so it does not overwrite the envloop SDK namespace.

After publication:

pip install envloop-cli==0.1.1
el --help
el train --help
el --api-url https://api.envloop.ai --workspace-id YOUR_WORKSPACE train job list

HTTP requests use system proxies by default (HTTPS_PROXY / HTTP_PROXY, with ALL_PROXY as a fallback); lowercase equivalents and NO_PROXY are honored. Without a configured proxy, requests connect directly.

Set ENVLOOP_API_KEY privately in your environment. Existing command names, arguments, JSON output, error codes and envloop/0.1 User-Agent are preserved. Local Task/Profile commands still require a separately installed compatible EnvPlatform, just as before; this package does not bundle an execution backend. No Tinker SDK, GPU, Docker or EnvHub dependencies are needed for remote queries.

Boundary and layout

el-cli/
  pyproject.toml
  src/el_cli/
    __init__.py
    cli.py
    client.py
    http.py
    train.py
    utils/{__init__,logging}.py
  tests/

The actual source has envloop/cli.py and envloop/train.py, rather than the cli/commands/train.py layout in the brief. cli.py imports Client and Train; Client uses the logging helper and lazily loads HttpService and Train. All use the standard library. Client also lazily imports EnvPlatform for local execution. Retaining Client preserves existing task/job/profile behavior rather than shipping a train-only CLI. These modules were extracted from the training-read CLI at b546830, based on main 11a5803. Shared SDK/transport fixes should be applied to both packages while their copied implementations remain separate.

The SDK keeps its Python APIs and legacy python -m envloop.cli interface, but no longer registers el; install this project for that executable.

Build, verify and publish

From this directory:

uv sync --locked
uv run --locked pytest -q --ignore=tests/test_jobs_http.py
uv run --locked ruff check .
uv build
python -m venv /tmp/el-cli-clean
/tmp/el-cli-clean/bin/python -m pip install dist/envloop_cli-0.1.1-py3-none-any.whl
/tmp/el-cli-clean/bin/el --help
/tmp/el-cli-clean/bin/el train --help

The real JobMaster integration cases in tests/test_jobs_http.py also require a separately installed EnvPlatform and its dependencies. Run the complete suite with that backend installed; the other cases need only the dev group.

The workstation's Aliyun index is a download mirror, not an upload endpoint. No existing private upload repository or publishing credentials were found on 2026-10-08. The intended public target is PyPI, upload endpoint https://upload.pypi.org/legacy/. Release remains unpublished until an authorized PyPI token (or configured Trusted Publisher) is available. Never put credentials in this project. With a token supplied privately as UV_PUBLISH_TOKEN:

uv publish --publish-url https://upload.pypi.org/legacy/ dist/*

Building a wheel does not establish that the PyPI name is reserved or that a release is available. Verify index installation after a successful upload.