Create Dataset on Cohere

Starts OAuth authorization for a connector via POST /v1/connectors/{id}/oauth/authorize. Returns redirect or auth payload per API. Use for secured data sources; unrelated to CO_TEST_CONNECTOR which revalidates via PATCH.

Deletes a fine-tuned model with DELETE /v1/finetuning/finetuned-models/{id}. Returns status. Use to retire custom models; destructive compared to CO_GET_FINETUNED_MODEL.

Lists models with GET /v1/models using page_size and page_token pagination plus optional endpoint filter. Returns model ids and capabilities. Use to discover IDs for CO_CHAT or CO_EMBED; paginate with next_page_token.

Updates fine-tuned model metadata via PATCH /v1/finetuning/finetuned-models/{id}. Returns updated record. Use for renames or settings; use CO_DELETE_FINETUNED_MODEL to remove.

Forces JSON object responses via response_format on POST /v2/chat. Returns message content as parseable JSON when the model complies plus usage metadata. Use for extraction, classification-as-JSON, or schemas; validate output. Use CO_CHAT when free-form prose is acceptable or CO_CHAT_WITH_TOOLS when calling APIs instead of JSON mode.

Reads organization dataset storage usage via GET /v1/datasets/usage. Returns quota consumption metrics for datasets and embed jobs storage. Use for capacity planning; for per-call token counts inspect chat response usage fields instead of this endpoint.

Cancels an embed job with POST /v1/embed-jobs/{id}/cancel. Returns updated job. Use to abort long batches; unlike CO_GET_EMBED_JOB this mutates state.

Lists embed jobs with GET /v1/embed-jobs and pagination. Returns job statuses. Use to monitor queues; use CO_GET_EMBED_JOB for one id.

Registers a connector with POST /v1/connectors; Cohere tests the endpoint during creation. Returns the new connector. Use for RAG search integrations; use CO_UPDATE_CONNECTOR to change URL or name.

Embeds document chunks with POST /v2/embed using input_type preset search_document. Returns embedding vectors aligned to input order for indexing pipelines. Use when building a vector store; use CO_EMBED_QUERY for user queries and CO_EMBED for custom input_type such as classification.

Creates an async batch job via POST /v2/batches with JSON body per Cohere batches API. Returns batch id. Use for high-volume offline processing; poll with CO_V2_GET_BATCH.

Fetches connector metadata with GET /v1/connectors/{id}. Returns URL and configuration. Use to verify registration; use CO_AUTHORIZE_CONNECTOR for OAuth flows.

Retrieves dataset details with GET /v1/datasets/{id}. Returns status, schema, and validation info. Use before launching embed jobs or fine-tunes; use CO_LIST_DATASETS to discover IDs.

Embeds a query string or batch via POST /v2/embed with input_type search_query. Returns vectors suited for similarity search against document embeddings. Use at retrieval time; pair with CO_EMBED_DOCUMENTS outputs in your vector DB. Use CO_EMBED if you need clustering or classification input types instead.

Starts fine-tuning with POST /v1/finetuning/finetuned-model body per API. Returns job or model handle. Use with CO_LIST_FINETUNING_EVENTS to track; use CO_CREATE_FINETUNED_MODEL_AND_WAIT to block until terminal state.

Creates embeddings via POST /v2/embed for texts or images with model, input_type, embedding_types, and truncate. Returns float/int8/uint8/binary vectors per item. Use input_type search_document when indexing corpora and search_query at query time. Prefer CO_EMBED_DOCUMENTS or CO_EMBED_QUERY when you want those presets without remembering enum values.

Deletes a dataset with DELETE /v1/datasets/{id}. Returns confirmation. Use to clean up obsolete corpora; irreversible unlike CO_GET_DATASET.

Creates a v2 batch then polls GET /v2/batches/{id} until terminal status or timeout. Returns final batch JSON. Use in CI; prefer CO_V2_CREATE_BATCH for manual monitoring.

Updates connector fields via PATCH /v1/connectors/{id}. Returns updated connector. Use to fix URLs; may trigger revalidation server-side. Use CO_DELETE_CONNECTOR to remove.

Lists connectors with GET /v1/connectors?limit&offset pagination. Returns connector ids ordered by recency. Use before CO_GET_CONNECTOR; use CO_CREATE_CONNECTOR to register a new search endpoint.

Deletes a connector with DELETE /v1/connectors/{id}. Returns confirmation. Use to unregister search endpoints; unlike CO_GET_CONNECTOR this removes the resource.

Calls POST /v2/chat with tools JSON parsed and tool_choice defaulted to auto so the model may emit tool_calls. Returns the same chat response envelope as CO_CHAT plus tool invocation payloads when applicable. Use for agent workflows that call your functions; use CO_CHAT when no tools are needed. For structured JSON without tools prefer CO_CHAT_JSON_MODE.

Reranks documents for a query using POST /v2/rerank with model, query, documents, top_n, return_documents. Returns ordered results with relevance scores. Use after vector recall to improve precision; use CO_RERANK_SIMPLE for a fast top-3 default. For single-score similarity without a query list use embeddings instead.

Revalidates connector configuration by PATCH /v1/connectors/{id} with an updated url (or name) so Cohere re-runs connectivity checks. Returns updated connector object. Use after changing infra; for read-only inspection use CO_GET_CONNECTOR instead.

Transcribes audio via POST /v2/audio/transcriptions with multipart or JSON per API. Returns text transcript. Use for speech input pipelines; unrelated to CO_CHAT text flows.

Runs Cohere chat completion on POST /v2/chat with full controls (model, messages, tools, documents, response_format, sampling, safety_mode, stream). Returns assistant text, citations, tool calls, token usage, and finish reason. Use as the primary chat entry point when you need every v2 parameter. Prefer CO_CHAT_JSON_MODE for strict JSON, CO_CHAT_STREAM when you must stream tokens, or CO_CHAT_WITH_TOOLS for agentic tool loops.

Creates a fine-tune then polls GET /v1/finetuning/finetuned-models/{id} until status is terminal or timeout. Returns final model payload. Use when scripts must block; prefer CO_CREATE_FINETUNED_MODEL for fire-and-forget.

Runs RAG-style chat on POST /v2/chat by supplying a documents JSON array alongside messages. Returns grounded answers with optional citations depending on model behavior. Use when you already have chunked documents from your datastore; use CO_EMBED plus a vector DB when you must retrieve docs dynamically first.

Classifies inputs with few-shot examples via POST /v1/classify. Returns predicted labels and confidences per input. Use for short text classification when you can supply labeled examples; for generative labeling use CO_CHAT_JSON_MODE instead.

Lists embed-compatible models via GET /v1/models?endpoint=embed. Returns models suitable for CO_EMBED. Use when building RAG; use CO_LIST_RERANK_MODELS for rerankers.

Creates a batch embed job via POST /v1/embed-jobs. Returns job id for polling. Use for large corpora; pair with CO_GET_EMBED_JOB or CO_CREATE_EMBED_JOB_AND_WAIT.

Lists training events for a model via GET /v1/finetuning/finetuned-models/{finetuned_model_id}/events. Returns chronological log lines. Use for debugging failed jobs; use CO_LIST_FINETUNING_CHECKPOINTS for step metrics.

Gets one model via GET /v1/models/{model}. Returns details for routing and deprecation flags. Use after CO_LIST_MODELS; for filtered lists use CO_LIST_CHAT_MODELS shortcuts.

Lists batches with GET /v2/batches and limit/after pagination when supported. Returns batch job summaries. Use to audit workloads; use CO_V2_GET_BATCH for detail.

Legacy text generation via POST /v1/generate with prompt-level controls (max_tokens, temperature, p, k, stop_sequences, return_likelihoods). Returns generations and likelihood metadata when requested. Use when integrating older Cohere examples; prefer CO_CHAT for conversational models and tool use.

Converts token IDs back to text via POST /v1/detokenize. Returns the joined string for debugging or inspection. Use after CO_TOKENIZE; not needed for normal chat flows.

Cancels an in-progress batch via POST /v2/batches/{id}:cancel per Cohere v2 batches API. Returns cancel confirmation payload. Use to stop runaway jobs; use CO_V2_GET_BATCH to confirm status afterward.

Reads training step metrics via GET /v1/finetuning/finetuned-models/{finetuned_model_id}/training-step-metrics (checkpoint-style curves). Returns metric series for loss and quality. Use alongside CO_LIST_FINETUNING_EVENTS; for model metadata alone use CO_GET_FINETUNED_MODEL.

Lists rerank-compatible models via GET /v1/models?endpoint=rerank. Returns rerank SKUs for CO_RERANK. Use when selecting rankers; use CO_LIST_MODELS for unfiltered catalogs.

Lists fine-tuned models with GET /v1/finetuning/finetuned-models and optional pagination params. Returns model ids and states. Use before CO_GET_FINETUNED_MODEL; use CO_CREATE_FINETUNED_MODEL to start training.

Fetches one fine-tuned model via GET /v1/finetuning/finetuned-models/{id}. Returns settings and status. Use to poll training completion; use CO_LIST_FINETUNING_EVENTS for detailed logs.

Sets stream=true on POST /v2/chat to receive newline-delimited streamed events (not SSE parsed here—raw body). Returns streamed chunks as returned by the API. Use for low-latency UX; for a single final string use CO_CHAT with stream false. For JSON guarantees combine streaming carefully with CO_CHAT_JSON_MODE semantics.

Gets embed job status via GET /v1/embed-jobs/{id}. Returns progress and output references. Use for polling; use CO_CANCEL_EMBED_JOB to stop.

Summarizes long text via POST /v1/summarize with length, format, extractiveness, and optional additional_command. Returns a concise summary string. Use for document digests; for chat-style Q&A use CO_CHAT_WITH_DOCS or CO_CHAT instead.

Gets one batch job via GET /v2/batches/{id}. Returns status and output handles. Use after CO_V2_CREATE_BATCH; use CO_V2_LIST_BATCHES to discover ids.

Creates an embed job then polls GET /v1/embed-jobs/{id} until complete or timeout. Returns final job record. Use in automation; prefer CO_CREATE_EMBED_JOB for interactive flows.

Calls POST /v2/rerank with top_n fixed to 3 unless overridden for quick triage of the best few chunks. Returns the top results and scores only. Use for fast UX or LLM context trimming; use CO_RERANK when you need custom top_n or return_documents toggles.

Lists datasets with GET /v1/datasets using limit and pagination before/after cursors as supported. Returns dataset metadata for training or eval. Use to browse corpora; use CO_GET_DATASET for one record. Paginate with next cursors from responses when present.

Validates the API key and returns metadata via GET /v1/check-api-key. Returns organization or key hints. Use for onboarding diagnostics; routine auth validation uses validateCredentials instead.

Tokenizes text via POST /v1/tokenize for a given model. Returns token IDs for budgeting or analysis. Use before estimating costs; use CO_DETOKENIZE to reconstruct text from IDs.

Lists chat-compatible models using GET /v1/models?endpoint=chat (and optional default_only). Returns paginated models for conversational use. Use to pick a CO_CHAT model; use CO_LIST_EMBED_MODELS for embedding SKUs.