openai.core — net.clojars.savya/openai-clj 0.4.0

Idiomatic Clojure wrapper over the official OpenAI Java SDK (com.openai/openai-java), focused on the Responses API.

cancel-response clj

(cancel-response client response-id)

Cancel an in-progress response by id and return the resulting response map.

client clj

(client)

(client {:keys [api-key organization project base-url timeout-ms max-retries
                azure-service-version]})

An OpenAI client. With no args, resolves credentials from the environment (OPENAI_API_KEY). Pass explicit config keys to set client options: :api-key, :organization, :project, :base-url, :timeout-ms, :max-retries, and :azure-service-version (an Azure OpenAI api-version string, used together with an Azure :base-url).

compact clj

(compact client response-id)

Compact a previous response by id and return the compacted response map.

count-input-tokens clj

(count-input-tokens client req)

Count input tokens for a Responses request shape. Accepts the same request map style as create-response; fields unsupported by the SDK's input token count endpoint are ignored. Returns {:input-tokens n}.

create-embeddings clj

(create-embeddings client req)

Create embeddings for :input (a string, or a vector of strings) with :model (required, e.g. "text-embedding-3-small"). Optional: :dimensions (truncated output size, supported by v3 models) and :user.

Returns {:model "..." :embeddings [[floats ...] ...] :usage {:prompt-tokens n :total-tokens n}}; :embeddings is ordered to match the input order (one vector even for string input).

create-response clj

(create-response client req)

Send a Responses API request and return a Clojure map.

Request keys: :model (required string), :input (required string or vector), :instructions, :max-output-tokens, :temperature, :top-p, :metadata, :previous-response-id, :store, :reasoning, :user, :tools, :tool-choice, :parallel-tool-calls, :background, :include, :truncation, :prompt-cache-key, :safety-identifier, :service-tier, :max-tool-calls, :top-logprobs, :json-schema, :verbosity, :conversation, :stream-options, and :moderation.

Message-vector input items accept {:role :system|:developer|:user|:assistant :content "..."}, multimodal content vectors containing text, image, or file part maps, and {:type :function-call-output :call-id "..." :output "..."}. Map outputs are JSON-encoded.

Structured outputs: :json-schema {:name "..." :schema {...} :strict true :description "..."}.

Tools: {:type :function :name "..." :description "..." :strict true :parameters {...}}, {:type :web-search}, {:type :file-search :vector-store-ids [...]}, or {:type :code-interpreter :container "..."}. Code interpreter defaults to an auto container when :container is omitted.

Tool choice: :auto, :required, :none, or {:type :function :name "..."}.

Returns {:id :model :status :output :text :usage :created-at} plus :error, :incomplete-details, or :previous-response-id when present. Output items are normalized to :message, :function-call, :reasoning, :web-search-call, :file-search-call, :code-interpreter-call, or :unknown.

delete-response clj

(delete-response client response-id)

Delete one stored response by id. The OpenAI Java SDK returns void.

get-model clj

(get-model client model-id)

Retrieve one model by id as a {:id :created :owned-by} map.

get-response clj

(get-response client response-id)

Retrieve one stored response by id as a response map.

list-input-items clj

(list-input-items client response-id)

List input items for a stored response id as normalized maps. Pages are followed automatically.

list-models clj

(list-models client)

List available models as a vector of {:id :created :owned-by} maps. Pages are followed automatically.

retrieve-streaming clj

(retrieve-streaming client response-id on-event)

Resume streaming an existing background response id. Invokes on-event with normalized event maps and returns concatenated output text, matching stream.

stream clj

(stream client req on-event)

Stream a Responses API request, invoking on-event with a normalized event map for every server-sent event as it arrives, and returning the concatenated output text. Takes the same req map as create-response. The underlying HTTP stream is closed automatically.

stream-text clj

(stream-text client req on-text)

Stream a Responses API request, calling on-text with each output text delta string as it arrives, and returning the full concatenated text.