polyfire-js

Provides pre-built React components that wrap LLM inference APIs, enabling developers to embed chat interfaces directly into React applications without building UI from scratch. Components handle message state management, streaming response rendering, and API integration through a declarative component API that abstracts away raw HTTP calls to language model endpoints.

Abstracts away provider-specific API differences (OpenAI, Anthropic, etc.) behind a unified interface, allowing developers to swap LLM providers or run requests against multiple providers without changing component code. Handles request normalization, response parsing, and error handling across different API schemas and authentication mechanisms.

Handles server-sent events (SSE) or chunked HTTP responses from LLM APIs, progressively rendering token-by-token output to the UI as it arrives rather than waiting for the complete response. Manages buffering, error recovery during streaming, and automatic UI re-renders on each token chunk using React's state update mechanisms.

Provides a templating system for constructing dynamic prompts with variable substitution, allowing developers to define reusable prompt patterns with placeholders that get filled at runtime from component props or user input. Supports conditional sections and formatting helpers to construct complex prompts without string concatenation.

Manages conversation history by storing messages in component state or external storage, automatically handling context window limits by truncating or summarizing older messages to fit within LLM token limits. Implements sliding window or summarization strategies to maintain conversation coherence while respecting model constraints.

Provides built-in error handling for API failures, network timeouts, and rate limiting, with configurable fallback strategies such as retry logic with exponential backoff, fallback to cached responses, or displaying user-friendly error messages. Distinguishes between recoverable errors (retry) and permanent failures (show error UI).

Provides TypeScript type definitions and runtime prop validation for all components, ensuring developers catch configuration errors at compile time and preventing runtime crashes from invalid props. Uses TypeScript interfaces and optional runtime schema validation to enforce correct component usage.

Exposes core functionality as React hooks (useChat, useCompletion, etc.) that can be composed into custom components, allowing developers to build their own UI while reusing the underlying LLM integration logic. Hooks manage state, API calls, and lifecycle independently of UI rendering.

Provides utilities to parse LLM responses into structured data (JSON, markdown tables, etc.), extracting specific fields or validating response format against a schema. Handles parsing errors gracefully and can re-prompt the LLM if the response doesn't match the expected format.