typescript-sdk_skill

Overview

This skill provides a complete reference for integrating with 300+ AI models using the OpenRouter TypeScript SDK and the callModel pattern. It documents authentication (API key and OAuth PKCE), input formats, streaming and synchronous response methods, tool creation, multi-turn flows, and security suggestions. The content is focused on practical code examples and patterns for production usage.

How this skill works

The SDK exposes a single, type-safe callModel function that accepts flexible inputs (strings, message arrays, multimodal content) and returns a result object with multiple consumption methods (getText, getResponse, getTextStream, etc.). It handles API key and OAuth PKCE authentication, automatic tool execution, streaming, and multi-turn conversations, while allowing dynamic parameters and stop conditions. Tools are defined with strong typing (Zod) and can be regular, generator, or manual, including hooks to modify next-turn parameters.

When to use it

• Implement text generation, summarization, or chat features across many model providers through a unified API.
• Build user-facing apps where users manage their own keys via OAuth PKCE.
• Execute external logic automatically during generation using typed tools (APIs, databases, web search).
• Stream responses to clients or consume full responses server-side with usage metadata.
• Run multi-turn workflows with stop conditions, cost limits, or tool-driven control flows.

Best practices

• Store API keys in environment variables and rotate keys regularly using the key management API.
• Use OAuth PKCE for user-owned keys so the app never handles user credentials directly.
• Define tools with Zod schemas for automatic input/output validation and safer execution.
• Prefer streaming (getTextStream) for low-latency UX and getResponse for auditing token usage and costs.
• Use stopWhen conditions and minimal-scoped keys to limit unexpected costs or runaway loops.

Example use cases

• Chatbot backed by multiple model providers selectable at runtime via callModel dynamic parameters.
• Search + analyze pipeline where a web_search generator tool streams progress events while the model composes results.
• User-authenticated apps: start OAuth flow, exchange code for a user API key, then callModel with the user's key.
• Background agents that call tools (calculators, databases) automatically and continue the conversation using nextTurnParams.
• Live streaming editor that displays tokens as they arrive using getTextStream for collaborative writing.