You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
Add typed/Pydantic models for OpenShell specs #1822
Constructing OpenShell specs directly from raw protobuf objects can be verbose and hard to validate. Typed Python models would improve developer experience and make configuration errors easier to understand.
Proposed change
Add typed models, potentially Pydantic-based, for common OpenShell spec construction. These models should validate user input and convert explicitly to the existing protobuf types.
Acceptance criteria
Add typed model(s) for common sandbox/session/spec construction paths.
Provide clear validation errors for invalid inputs.
Keep protobuf interop explicit and backward compatible.
Add unit tests for valid construction, invalid values, and protobuf conversion.
Document at least one end-to-end example.
Notes
This is primarily a Python SDK developer-experience improvement and should not require gateway/proto changes.
@NVIDIA/openshell-codeowners, could you please help with triage on this SDK request?
Would typed, validated Python models for common OpenShell specs be useful as part of the supported SDK surface? If so, may I revise my experimental branch to the preferred model strategy and submit it as a PR?
The current spike uses dataclasses, while this issue mentions Pydantic-style validation. I would appreciate direction on that choice and on whether RFC-0008 (#1764), #2565, or another SDK effort already owns these public models.
Activity
dhirajsb commentedon Aug 4, 2026
@NVIDIA/openshell-codeowners, could you please help with triage on this SDK request?
Would typed, validated Python models for common OpenShell specs be useful as part of the supported SDK surface? If so, may I revise my experimental branch to the preferred model strategy and submit it as a PR?
The current spike uses dataclasses, while this issue mentions Pydantic-style validation. I would appreciate direction on that choice and on whether RFC-0008 (#1764), #2565, or another SDK effort already owns these public models.