A JSON parser that can parse valid json from a stream and produce a safe-to-parse result based on the schema provided. This allows for access to the data while it is being parsed.

Structured extraction in Typescript, powered by llms, designed for simplicity, transparency, and control.

A universal LLM client - extends the official openai sdk to provide support for providers that do not adhere to the same api and format, like Anthropic or Azure. One universal sdk for all the top LLMs from Together, OpenAI, Microsoft, Anyscale and Anthropic

Define structured response models for OpenAI or Anyscale completions using Zod schemas and enable partial streaming of that json so that it can be used safely and right away.

A collection of React hooks for working with streams, but primarily designed to work with LLM response streams.

A dynamic prompt enhancer cli - given a prompt, set of criteria and optional dataset, Edna will generate a series of evals and generate a new higher scoring prompt given your criteria.

A tree of thought chat expirement. Given a single question the program will generate a recursive tree of scored and dynamically weighted thoughts, questions and answers--ultimately converging back on a final answer..

Open source LLM proxy & observability.

Evals in Typescript

Quick OpenAI assistants with Next.js - will create all of the api routes you need with one command. And provides a useThread hook for interaction with the assistant through those endpoints.

A tool for creating a parent iframe for managing a chat widget install from Drift.com - this enables enterprises with strict CSP policies to install the chat widget without having to loosen their CSP policy.