v0.5.0: deploy to agent_runtime builds on the CLI's own Python (sys.version_info), not the project's — fails when a dependency lacks a wheel for that version (e.g. Python 3.14 + litellm)

[qingvincentyin]

Summary

agents-cli deploy to --deployment-target agent_runtime (Vertex AI Agent Engine) determines the build/runtime Python version from the interpreter that runs the CLI itself, not from the target project's Python. When agents-cli is installed as a uv tool under a Python that some required dependency doesn't support yet, the server-side build fails with only a generic error. Concretely, with the CLI running on Python 3.14, the build fails because litellm (a transitive dependency of google-adk) has no 3.14-compatible release. This also builds on a Python version that the generated project's own pyproject.toml requires-python = ">=3.11,<3.14" excludes.

Environment

• agents-cli v0.5.0 (installed as a uv tool; uv 0.11.14)
• CLI tool environment interpreter: Python 3.14.5 (machine default)
• Project: scaffolded adk template, --deployment-target agent_runtime, requires-python = ">=3.11,<3.14"

Repro

1. Have agents-cli installed under Python 3.14 (e.g. uv tool install google-agents-cli on a machine whose default Python is 3.14).
2. Scaffold an adk agent targeting agent_runtime.
3. agents-cli deploy --project <PROJECT_ID> --update-env-vars "GOOGLE_CLOUD_LOCATION=global,GOOGLE_GENAI_USE_VERTEXAI=True"

Actual

Error: Deployment failed: {'code': 3, 'message': 'Build failed. The issue might be caused by incorrect code, requirements.txt file or other dependencies. ...'}

The real error is only visible server-side in Cloud Logging (resource.type="aiplatform.googleapis.com/ReasoningEngine"):

ERROR: Ignored the following versions that require a different python version: ... 1.85.4 Requires-Python >=3.10,<3.14 ...
ERROR: Could not find a version that satisfies the requirement litellm==1.85.4
ERROR: No matching distribution found for litellm==1.85.4

Build install paths confirm the builder ran on 3.14 (.venv/lib/python3.14/..., cp314 wheels).

Expected

The deploy should build on a Python compatible with the project (its .venv / requires-python), or fail fast with a clear message instead of a generic server-side build failure.

Root cause

The Agent Engine SDK derives the build version from the calling interpreter when no override is passed — vertexai/_genai/agent_engines.py:

if python_version:
    sys_version = python_version
else:
    sys_version = f"{sys.version_info.major}.{sys.version_info.minor}"

agents-cli does not pass python_version, so the build inherits the CLI's own interpreter (3.14). There is no --python-version flag to override it.

Likely a regression in v0.5.0

The same project deployed successfully on agents-cli v0.3.0, with the CLI on the same Python 3.14 tool environment, even though litellm has never supported 3.14. That implies v0.3.0 derived the build Python from the project (≤3.13), and v0.5.0 changed to the sys.version_info fallback. (Stated as an inference from observed behavior, not a code diff.)

Suggested fixes (in order of preference)

1. Set the deploy python_version from the Python the requirements were resolved against (the project .venv / uv.lock).
2. Otherwise honor the project's requires-python upper bound (don't build on an excluded version).
3. At minimum, add a --python-version flag and warn when the build Python differs from the resolution Python.

Workaround

Run the CLI under a supported Python: uvx --python 3.13 --from google-agents-cli agents-cli deploy ..., or durably uv tool install google-agents-cli --python 3.13.