Skip to content
providerkit
GitHub

Call models. Run agents.
Build chat screens.

Call a provider with Core. Add Agents for tools and workflows, or React for a chat screen. Your app checks permissions and stores the data.

Choose where to startFind an example

Try a task

Would waiting help?

A Core example. Choose a failure to run the real function in your browser.

Call

isTransient("overload")

Import and setup

Result

true

This kind of failure can pass with time.

Retry, or fall back to another model.

No API key or network call. A transient kind is not permission to replay a request. Read the retry rules.

Try the full error classifier

Core

@providerkit/core

Call a model, stream its answer and handle failures. Keep your own agent loop, or use ours.

0.35.1. Chat, hosted tools, speech, transcription and opt-in tracing. The main entry has no runtime dependencies.

Make your first callExplore Core

Agents

@providerkit/agents

Give an agent a task. It calls a model, runs the tools you allow and returns a result you can check.

0.4.0, experimental. Requests and saved steps share the same execution limits. Your host owns storage.

Run an agent offlineExplore Agents

React

@providerkit/react

Chat state, hooks and optional components for an app’s conversation screen.

0.2.0, experimental. Use hooks or plain-data components with your app’s transport and styles, or pair with Native 0.1.0 on phones.

Add a chat screenTry the components

Keep the parts you already have.

Use Core under your own loop. Add Agents when you want a shared loop. Your app keeps sign-in, permission checks and stored conversations.

Understand the boundariesLook up an API

Reading with an AI assistant? llms.txt indexes the docs. Each page has a Markdown version, and kinds.json lists error remedies as data. All packages are MIT licensed.