Migrate sample to OGX (Open GenAI Stack) 1.2.5

[JslYoon]

Summary

Migrates this sample from Llama Stack to OGX (Open GenAI Stack) 1.2.5, the upstream rename of Llama Stack. Covers the Python app SDK, the server runtime config, the Backstage/RHDH software template, k8s resource names, and docs.

https://redhat.atlassian.net/browse/RHIDP-15863

https://redhat.atlassian.net/browse/RHIDP-17526

What changed

App SDK — llama-stack-client/llama-stack replaced with ogx-client==1.2.5:

• OgxClient(base_url=...); list endpoints now return a response object with a .data list (vector_stores.list().data, vector_stores.files.list(...).data, models.list().data) instead of being directly iterable.
• Create endpoints take request objects: vector_stores.create(OpenAICreateVectorStoreRequestWithExtraBody(name=...)), vector_stores.files.create(open_ai_attach_file_request=OpenAIAttachFileRequest(file_id=...)).
• files.create(file=<str path>, purpose=...) takes a str path (preserves basename).
• Type remaps aliased to old names to minimize churn (OpenAIResponseObject, OpenAIFileObject, OpenAIResponseOutputMessageFileSearchToolCallResults).
• Added explicit openai>=2.14.0,<3 (previously a transitive dep of llama-stack-client; pinned <3 because v3 drifts the beta.chat.completions.parse surface the code uses).

Server + template + infra — rebranded to OGX:

• Server deps/config under start-local-ogx/ (was start-local-llamastack/); run.yaml migrated to the OGX 1.2.5 API surface.
• Template (ogx-agentic), secret ogx-secrets, k8s resources (${name}-ogx, service-ogx), runtime sidecar image agentic-ogx.
• Docs updated.

Fixes bundled in:

• Safety check runs via Llama Guard chat instead of /v1/moderations.
• unknown classification routing renders a graceful prompt instead of an error.
• Server starts with --insecure (no TLS certs required locally).

Kept intentionally: env var names LLAMA_STACK_URL / LLAMA_STACK_SERVER_OPENAI (their values point at the renamed -ogx service), config key llamastack:, check_llama_stack_availability, and ollama/llama-guard/.llama data dirs. Repo name unchanged.

Testing

• 180 unit tests pass; ruff clean.
• Note: unit tests are mocked and do not prove runtime correctness against a live OGX server — a live smoke test is still recommended before merge.

Note on images

Template appContainer/ogxContainer point at quay.io/redhat-ai-dev/*. redhat-ai-dev/agentic-ogx:latest is not published yet — upstream CI needs to build/publish it for deploys to pull the OGX runtime sidecar.

🤖 Generated with Claude Code

[gabemontero]

use of a local ogx instance is not working (the instance does not come up) ... I've reported it to @JslYoon and he is working it

in the interim I'm pivoting to deploying on OCP and giving that a go

[gabemontero]

so @JslYoon the ogx deployment is failing when I launch the template because it tries to pull the image

quay.io/redhat-ai-dev/agentic-ogx:latest

which does not exist

looking at https://quay.io/organization/redhat-ai-dev there is an agentic-llama-stack:latest

IIRC @maysunfaisal did work to get the agentic-llama-stack image out there ... now, you probably want a newer image for OGX, so there is more involved than just retagging / renaming that image