cohere-java-sdk_skill

Overview

This skill is a practical reference for using the Cohere Java SDK to implement chat, streaming, embeddings, reranking, and tool-enabled agents in Java or Kotlin apps. It summarizes client setup, key API patterns, and critical behaviors (like embedding input types and streaming handling) so you can integrate Cohere APIs reliably.

How this skill works

The skill shows how to configure the Cohere client (env var, explicit key, or custom endpoint) and demonstrates common API flows: synchronous chat, streaming chat events, embeddings for storage vs query, reranking candidate documents, and tool/function invocation. It includes example request/response handling and a loop pattern for executing tool calls until the assistant returns a final answer. Error handling examples show how to catch API errors and general exceptions.

When to use it

• Building chatbots or assistants in Java/Kotlin that call Cohere chat models
• Implementing streaming responses to display tokens or deltas in real time
• Creating vector search pipelines where embeddings are stored and queried
• Re-ranking candidate documents or answers before presenting results
• Implementing agent/tool patterns where the model calls application functions

Best practices

• Initialize the Cohere client once and reuse it across requests to avoid overhead
• Set embedding inputType explicitly: use SEARCH_DOCUMENT for stored docs and SEARCH_QUERY for queries
• Use streaming when you need low-latency token-by-token UI updates and process content deltas carefully
• Handle tool calls in a loop: add assistant/tool messages back into the context and re-call chat until no toolCalls remain
• Set sensible sampling and guardrail parameters (temperature, maxTokens, stopSequences) for predictable outputs
• Catch CohereApiException separately to surface API status codes and messages

Example use cases

• A Kotlin FAQ bot that embeds support articles with SEARCH_DOCUMENT and answers queries by embedding SEARCH_QUERY and reranking passages
• A live coding assistant that streams partial responses into an IDE plugin using chatStream
• A customer service agent that uses tools to fetch order status via defined functions and returns structured results
• A search augmentation pipeline that re-ranks initial retrieval results with rerank-v4.0-pro before returning top N
• A data enrichment job that batches documents to compute and store embed-v4.0 vectors for later semantic search