Summary

• add an optional Atlas Cloud engine backed by the OpenAI-compatible chat completions API
• route atlascloud/ model strings without changing existing engine defaults
• document credentials, endpoint configuration, and model selection

Validation

• python -m py_compile octotools/engine/atlascloud.py octotools/engine/factory.py
• focused mocked factory and chat-completions request test
• git diff --check
• one live Atlas Cloud chat-completions smoke request using a model from the current public catalog

The Atlas Cloud client sets max_retries=0, so a generation request is never submitted more than once by the SDK.