OpenAI

Source: https://runstack.engineer/docs/toolkits/openai

OpenAI models for text, images, audio, embeddings, and retrieval: run responses, transcribe speech, generate media, manage files, batches, fine-tuning, and vector stores.

Tools (46):

Tool	Description
`openaiCreateResponse`	Run the OpenAI Responses API: send a prompt (or full conversation input) to a model and get a model-generated answer. Supports instructions, built-in tools (web search, file search), function tools, JSON output format, and continuing a previous response.
`openaiGetResponse`	Retrieve a previously stored model response by ID, including its output items, usage, and status (useful for background responses or auditing).
`openaiListResponseInputItems`	List the input items of a stored response (the conversation/messages it saw), with paging via the returned next cursor.
`openaiCancelResponse`	Cancel a background (stream:false + background:true) response that is still running, so it stops consuming tokens.
`openaiDeleteResponse`	Delete a stored response and its server-side state permanently. Use when a response should no longer be retrievable.
`openaiCreateChatCompletion`	Create a chat completion via the classic Chat Completions API: a list of chat messages in, one model message out. Use when a consumer or library expects /chat/completions instead of the Responses API.
`openaiListModels`	List the models available to the connected API key (including fine-tuned models). Use to discover exact model IDs before calling inference tools.
`openaiGetModel`	Retrieve one model by ID: owner, display name, creation date, and the modalities/capabilities it supports.
`openaiDeleteModel`	Delete a fine-tuned model permanently. Base models cannot be deleted; the API returns an error for them.
`openaiCreateEmbedding`	Convert text into a numeric embedding vector for semantic search, clustering, or similarity. Send one string or a batch of strings in one call.
`openaiCreateModeration`	Classify text (or an image URL) for harmful content: hate, harassment, violence, self-harm, sexual, and illicit categories with per-category flags. Use to pre-screen user input before it reaches a model.
`openaiGenerateImage`	Generate an image from a text prompt with a GPT image or DALL-E model. Returns base64 image data (GPT image models) or a URL (dall-e models).
`openaiEditImage`	Edit or extend an image with a prompt: retouch, remove/replace objects, add elements, or restyle. Send the source image (and optionally a transparency mask) as base64; GPT image models accept up to 16 source images.
`openaiCreateSpeech`	Text-to-speech: synthesize spoken audio from up to 4096 characters of text. Returns the audio file as base64 that you can write to disk or play.
`openaiCreateTranscription`	Speech-to-text: transcribe an audio file (mp3, wav, m4a, ...) into text in its source language, optionally with word timestamps and speaker diarization.
`openaiCreateTranslation`	Translate an audio file into English text (speech-to-English-text). Use for foreign-language recordings when you only need the English transcript.
`openaiUploadFile`	Upload a file to OpenAI (JSONL training data, PDFs, text) so it can be used for fine-tuning, batches, or the Files/vector-store tools. Purpose examples: fine-tune, batch, vision, user_data, assistants.
`openaiListFiles`	List files uploaded to the OpenAI project, newest first, with purpose, size, and status (processed, uploading, error).
`openaiGetFile`	Retrieve file metadata: name, purpose, size, creation date, and processing status (check this before using a file in a batch or fine-tune).
`openaiGetFileContent`	Download the content of a text/JSONL file as a string (for transcripts, batch outputs, fine-tune datasets). Binary files come back base64.
`openaiDeleteFile`	Delete an uploaded file permanently from the OpenAI project.
`openaiCreateBatch`	Submit a batch of requests for asynchronous, lower-cost processing. Provide a JSONL file (one API request per line, uploaded with purpose "batch") and the endpoint those requests target (e.g. /v1/responses).
`openaiGetBatch`	Get a batch by ID: status (validating, in_progress, completed, failed, expired), request counts, output file IDs, and error counts.
`openaiListBatches`	List batches with their statuses; filter by endpoint and page through with the returned next cursor. Poll a batch here until it completes, then read its output file.
`openaiCancelBatch`	Cancel a batch that has not finished yet; completed requests stay available, pending ones are dropped.
`openaiCreateFineTuningJob`	Start a fine-tuning job on an uploaded training file (JSONL). Returns a job you poll with openaiGetFineTuningJob; the resulting model ID appears when status is succeeded.
`openaiListFineTuningJobs`	List fine-tuning jobs with their statuses (validating_files, running, succeeded, failed, cancelled) and page through results.
`openaiGetFineTuningJob`	Get one fine-tuning job: status, trained tokens, result files, error details, and the fine-tuned model ID once it succeeds.
`openaiListFineTuningEvents`	List the event log of a fine-tuning job (progress, metrics, errors) — the best way to see why a job is slow or failing.
`openaiCancelFineTuningJob`	Cancel a queued or running fine-tuning job so it stops consuming training budget.
`openaiPauseFineTuningJob`	Pause a running fine-tuning job; it can be resumed later with openaiResumeFineTuningJob.
`openaiResumeFineTuningJob`	Resume a paused fine-tuning job from where it left off.
`openaiCreateVectorStore`	Create an empty vector store (named collection of embedded files) for the file_search tool. Attach files afterwards with openaiAddVectorStoreFile or openaiCreateVectorStoreFileBatch.
`openaiListVectorStores`	List vector stores with file counts and statuses, paged with the returned next cursor.
`openaiGetVectorStore`	Get one vector store: name, file/incomplete-file counts, expiry, and status.
`openaiUpdateVectorStore`	Rename a vector store or change its expiry policy and metadata.
`openaiDeleteVectorStore`	Delete a vector store and its contents permanently (the attached files stay in Files).
`openaiSearchVectorStore`	Semantic search over a vector store: return the chunks of embedded files most relevant to a query (what the file_search tool does at runtime), with optional attribute filters.
`openaiAddVectorStoreFile`	Attach an uploaded file to a vector store so it gets embedded and becomes searchable (one file at a time).
`openaiListVectorStoreFiles`	List the files in a vector store with their embedding statuses (in_progress, completed, failed) — use to find failed ingestions.
`openaiDeleteVectorStoreFile`	Detach a file from a vector store (its vectors are removed; the file itself remains in Files).
`openaiCreateVectorStoreFileBatch`	Attach many files to a vector store in one embedding batch (much cheaper than attaching files one by one). Poll with openaiGetVectorStoreFileBatch.
`openaiGetVectorStoreFileBatch`	Get a file-embedding batch: overall status plus per-file counts (completed, in_progress, failed) for a vector store.
`openaiCancelVectorStoreFileBatch`	Cancel an in-progress file-embedding batch on a vector store.
`openaiGetUsageCompletions`	Token usage and request counts over time for the organization, bucketed by time and optionally grouped by model, project, or API key. Requires an ADMIN API key (a regular project key returns 401).
`openaiGetOrganizationCosts`	Billed cost in USD over time for the organization (line-item level spend). Requires an ADMIN API key (a regular project key returns 401).