DOCSEARCHbyAlgolia
Submit Docs

Cohere Documentation | LLMs, API, and SDKs

website·Indexed Apr 14, 2026

https://docs.cohere.com/

Official documentation for Cohere, a platform for enterprise AI and natural language processing. Includes guides for the v2 API, SDKs for Python, TypeScript, Java, and Go, and model families like Command R+, Command A, Aya, Embed, and Rerank. Covers use cases such as RAG, semantic search, tool use, audio transcription, and multimodal vision. Features integration details for cloud and private deployments on AWS, Azure, and on-premise environments.

Tokens
1.1M
Snippets
701
Records
7K
Agent score
50%

What's inside Cohere

  1. Introduction to Cohere Embeddings

    v1
    Embeddings represent the meaning of text, images, or mixed content as a list of numbers. By comparing these vectors using a similarity function, you can determine how related two pieces of information are. Common use cases include semantic search, clustering, and classification. To use embeddings, you typically generate vectors for input data and compare them against other vectors to find similar content.

  2. Introduction to Cohere Embeddings

    v2
    Embeddings represent the meaning of text, images, or mixed content as a list of numbers. By comparing these vectors using a similarity function, you can determine how related two pieces of information are. Common use cases include semantic search, clustering, and classification. To use embeddings, select a model (e.g., embed-v4.0) and generate vectors for your inputs.

  3. Overview of Cohere Reranking

    Reranking is a technique that provides a semantic boost to search quality for keyword (lexical) or vector (semantic) search systems. It is particularly useful in Retrieval Augmented Generation (RAG) workflows. The typical architecture involves a first-stage retrieval using an existing system (lexical or semantic) followed by a second-stage reranking using the Cohere Rerank endpoint to optimize the relevance of the returned results.

  4. Overview of Advanced Generation Parameters

    v1

    Cohere models support several advanced generation parameters that control how output tokens are selected. These parameters allow you to balance between deterministic and diverse text generation. The key parameters are:

    • top_p: Controls nucleus sampling by limiting the selection to tokens whose cumulative probability exceeds a threshold.
    • top_k: Limits the selection to the top K most likely tokens.
    • frequency_penalty: Reduces the likelihood of tokens that have already appeared in the generation.
    • presence_penalty: Reduces the likelihood of tokens that have appeared at least once in the generation.

    These parameters are applied after the model calculates likelihood scores for every token in its vocabulary.

  5. Overview: Building a QA Bot from Technical Documentation

    v1

    This guide demonstrates how to build a single-turn chatbot that answers user questions based on technical documentation. The workflow involves indexing documentation into a vector database, building a high-accuracy retriever using Cohere's rerank API, generating answers for a set of 100 evaluation questions, and comparing model outputs against golden reference answers to assess performance.

    The example uses the aws-documentation dataset, which contains over 26,000 AWS documentation pages preprocessed into 120,000+ chunks, paired with 100 real user questions.

  6. Overview of Cohere Foundational Models

    Cohere's foundational models are divided into two categories: generative and representative models. Generative models take a text input (prompt) and generate text output (response), suitable for use cases like question-answering, copywriting, and summarization. Representative models take a text input and generate a numerical representation (embedding) that captures contextual meaning, suitable for semantic search, text classification, and topic modeling.

  7. Introduction to Aya Vision

    v1
    Aya Vision is a state-of-the-art open-weights multimodal multilingual model developed by Cohere. It processes both text and image inputs to generate text responses. Key capabilities include question answering, multilingual understanding, image captioning, text recognition (OCR), classification, and comparative analysis of multiple images. The model is available as Aya Vision 32B on Hugging Face.

  8. Overview of Chroma and Cohere Integration

    v1
    Chroma is an open-source vector search engine designed for quick installation and immediate use with Python or JavaScript. This integration allows developers to build Retrieval-Augmented Generation (RAG) applications by combining Cohere's embedding models with Chroma's vector storage capabilities. The workflow involves generating embeddings via the Cohere API and storing them in Chroma for semantic search.

  9. What is a Private Deployment?

    A Private Deployment allows organizations to run Cohere AI models within a controlled, internal environment where the organization manages the deployment infrastructure. Cohere provides guidance and support for hardware/driver compatibility and container prerequisites. These deployments typically run on Kubernetes but do not strictly require it. There are two deployment types:

    1. On-premises (on-prem): You procure your own GPUs, servers, and hardware, retaining full control over data and the AI system on your own premises.
    2. Cloud (VPC): You host models on a cloud provider (AWS, Azure, GCP, or OCI) within a Virtual Private Cloud (VPC). Cohere supports any VPC on any cloud environment provided hardware requirements are met.

    In both scenarios, data never leaves your environment, and the model can be fully network-isolated.

  10. Overview of Command A Translate capabilities

    v2
    Command A Translate is Cohere's state-of-the-art machine translation model. It supports 23 languages and delivers industry-leading performance for translation tasks. For enterprise users, it offers full data control through private deployment options. This model is designed to handle automated translation from one language to another using LLM capabilities.

  11. Overview: Generating Multi-Faceted Queries for RAG

    This guide details how to build a tool use system using the command-a-03-2025 model to generate multi-faceted queries. This approach is designed for Retrieval Augmented Generation (RAG) systems that need to capture full user intent by leveraging structured metadata (e.g., programming language, creation time, tech stack) rather than passing raw queries directly to a search function. By decomposing user requests into specific facets, the system can filter large databases more precisely to retrieve relevant results.

  12. Overview of Command A Translate capabilities

    Command A Translate is Cohere's state-of-the-art machine translation model. It supports translation across 23 languages and delivers industry-leading performance. For enterprise users, it offers full data control through private deployment options. This model is designed to handle automated translation tasks effectively using large language models.