[jamesilse-solo]

Reframes the standalone and Kubernetes OpenAI SDK pages as guides to using SDKs with agentgateway, per the Slack thread on SDK golden paths.

Changes

• Retitle the page to "SDKs (Python, Node.js, Go)". The URL stays openai-sdk/.
• Intro explains that the OpenAI SDK reaches any configured provider; agentgateway translates where needed.
• Add a Go example (openai-go/v3).
• Add "Call more than one provider" with the openai/* and anthropic/* provider-prefix pattern.
• Add "Migrate from the LiteLLM SDK": same provider/model strings, before/after code, and a call mapping table.
• Add "Authenticate with a JWT": pass the IdP token as the SDK API key.
• Link the LiteLLM config import guide (main, latest) to the new migration section.

Kubernetes page:

• Fix the base URL note. It said to leave /v1 out of base_url; the SDK appends only /chat/completions, so requests returned 404 with the default route and with the /openai custom route. The page now says to end the base URL with /v1.
• Add the same Go, multi-provider, LiteLLM, and JWT sections, using AgentgatewayModel and AgentgatewayPolicy.
• Add a warning: after a listener allows AgentgatewayModel, it serves /v1/chat/completions from models, so an unprefixed model name that an HTTPRoute used to serve returns model not found. HTTPRoutes on other paths, such as /openai, keep working.

The multi-provider and LiteLLM sections render for 1.6.x and later only.

Testing

Ran against agentgateway 1.7.0 and 1.6.0-alpha.2 with the same results.

The page's Python, Node.js (openai 7.52.0), and Go (openai-go/v3 v3.96.0) snippets, run as written against the page's catch-all config: OpenAI returned responses for all three.

With jwtAuth (strict) and the provider-prefix config:

• Python sync, streaming, and AsyncOpenAI; Go openai-go/v3: OpenAI returned responses for openai/gpt-4o-mini
• anthropic/claude-sonnet-4-5: routed to api.anthropic.com with the prefix stripped (no Anthropic key, so it stopped at Anthropic auth)
• JWT passed as api_key: accepted, jwt.sub in access logs; missing or bad token returns 401

Kubernetes: agentgateway 1.7.0 on kind.

• The page's Python, Node.js, and Go snippets, run as written with INGRESS_GW_ADDRESS set to a port-forward: OpenAI returned responses for all three.
• Base URL without /v1: 404 on the default route and the /openai route. With /v1 (or /openai/v1): 200.
• AgentgatewayModel with openai/* and anthropic/*, plus a strict jwtAuthentication policy: same results as standalone for Python sync, streaming, async, and Go.
• Models and the default HTTPRoute on one listener: prefixed models work, unprefixed gpt-4o-mini returns model not found, /openai/v1 still works.

Not covered: Kubernetes 1.6.x, and page titles for 1.6.x and older.