Migrating From wkok/openai-clojure

wkok/openai-clojure wraps the Chat Completions API over hand-rolled HTTP; openai-clj wraps the official com.openai/openai-java SDK and the Responses API. The structural differences:

• Explicit client. wkok reads OPENAI_API_KEY ambiently and takes credentials in a per-call options map. openai-clj builds a client once ((openai/client) also reads OPENAI_API_KEY) and passes it as the first argument to every function.
• Responses, not Chat Completions. :messages becomes :input (a plain string or a vector of message items); the :role "system" message becomes :instructions; multi-turn state can use :previous-response-id instead of resending history.
• Flat function tools. wkok nests {:type "function" :function {:name ...}}; openai-clj flattens to {:type :function :name ... :parameters ...}, and tool call results go back as :function-call-output input items.
• Streaming is a function, not a flag. wkok's :stream true + :on-next becomes (openai/stream client params callback) or stream-text for text deltas only.
• Errors are ex-info. All failures carry :openai/error in ex-data (see Errors) instead of raw HTTP client exceptions.

wkok/openai-clojure	openai-clj
(api/create-chat-completion {:model m :messages [...]})	(openai/create-response client {:model m :input "..."})
{:messages [{:role "system" :content s} ...]}	{:instructions s :input [...]}
(api/create-chat-completion {...} {:api-key k :organization o})	(openai/client {:api-key k :organization o}) once, then pass client
{...} {:request {:timeout 60000}}	(openai/client {:timeout-ms 60000})
{:stream true :on-next f}	(openai/stream client params f) / (openai/stream-text client params f)
(api/create-embedding {:model m :input x})	(openai/create-embeddings client {:model m :input x})
(api/list-models) / (api/retrieve-model "id")	(openai/list-models client) / (openai/get-model client "id")
(api/upload-file {:purpose p :file f})	(openai/upload-file client {:purpose p :file f})
(api/retrieve-file-content "id")	(openai/file-content client "id") → byte[]
(api/create-batch {...}) / retrieve-batch / cancel-batch	create-batch / get-batch / cancel-batch, client-first
Azure: {:impl :azure} + env vars	(openai/client {:base-url "https://<resource>.openai.azure.com" :azure-service-version "..."})

Not covered here: images, audio, moderations (as a standalone API), assistants/threads/runs, and vector-store management are out of openai-clj's scope - keep wkok/openai-clojure for those surfaces or drop to the Java SDK directly.