feat: add Atlas Cloud backend helper

[binyangzhu000-sudo]

Pull Request

Issue

N/A

Description

Adds an Atlas Cloud OpenAI-compatible backend helper that reuses the existing OpenAIBackend path while applying Atlas Cloud defaults:

• adds AtlasCloudBackend and create_atlascloud_openai_backend
• reads ATLASCLOUD_API_KEY for auth, with optional ATLASCLOUD_API_BASE / ATLASCLOUD_BASE_URL overrides
• sets https://api.atlascloud.ai/v1 and qwen/qwen3.5-flash as defaults
• registers atlascloud, atlas-cloud, and atlas backend aliases for start_backend / start_session
• keeps provider metadata as atlascloud for telemetry paths

Testing

• Tests added to the respective file if code was changed
• New code has 100% coverage if code was added
• Ensure existing tests and github automation passes (a maintainer will kick off the github automation when the rest of the PR is populated)

Local validation:

• uv run ruff format mellea/backends/atlascloud.py mellea/stdlib/start_backend.py mellea/stdlib/session.py test/backends/test_atlascloud.py
• uv run pytest test/backends/test_atlascloud.py test/stdlib/test_session_unit.py -q
• uv run ruff check mellea/backends/atlascloud.py mellea/stdlib/start_backend.py mellea/stdlib/session.py test/backends/test_atlascloud.py test/stdlib/test_session_unit.py
• git diff --check
• uv run python -m py_compile mellea/backends/atlascloud.py mellea/stdlib/start_backend.py mellea/stdlib/session.py test/backends/test_atlascloud.py
• uv run mypy mellea/backends/atlascloud.py mellea/stdlib/start_backend.py mellea/stdlib/session.py test/backends/test_atlascloud.py
• uv run python tooling/docs-autogen/build.py
• uv run python tooling/docs-autogen/audit_coverage.py --docs-dir docs/docs/api --quality --fail-on-quality --threshold 100

The Atlas Cloud model catalog was checked live before committing; qwen/qwen3.5-flash and deepseek-ai/deepseek-v4-pro were both present.

Attribution

• AI coding assistants used

---

Adding a new component, requirement, sampling strategy, or tool?

• Component
• Requirement
• Sampling Strategy
• Tool

[planetf1]

@binyangzhu000-sudo thanks for the contribution.

Looking at the change you made to use Atlas Cloud as the backend, it seems as if this should already work w today with OpenAIBackend by passing base_url="https://api.atlascloud.ai/v1" and api_key=... directly (or via OPENAI_BASE_URL / OPENAI_API_KEY). I don't have access though to try this myself.

A new backend would make sense if it needs to implement new behaviour including there being limitations in the existing backends.

Was there a particular reason you looked at adding a backend - for extra function now? plans for the future? bugs in existing backend? poor docs?

[binyangzhu000-sudo]

You’re right that Atlas Cloud requests already work through OpenAIBackend by passing base_url and api_key; this PR does not add a new wire-protocol capability.

The intended value is provider-specific convenience: separate ATLASCLOUD_* environment variables (so users do not have to repurpose OPENAI_*), endpoint/model defaults, start_backend / start_session aliases, and an atlascloud telemetry provider label.

If provider-specific wrappers are not a direction the project wants, I’m happy to narrow this to documentation/a small helper or close it. Please let me know which scope you prefer.

[psschwei]

Thanks for the PR @binyangzhu000-sudo!

One of the things that your PR has surfaced is that adding a new provider requires editing core stdlib internals. While you do all of that correctly, that's not a pattern we want to perpetuate. So before we can merge a provider like this, we'd like to do the groundwork to make it easier for folks to add backends without needing to touch the core Mellea internals.

I'm opening an issue to spec that out and will link it here. Once we get that in, then we can resume the conversation about adding a new backend and/or documenting it.

Is that something you'd be interested in taking on?