Rebyte Agent Toolkit

Configure managed Rebyte Agents and add them to React applications.

Rebyte does not ship a second Responses client. Server applications execute an Agent with the official OpenAI SDK. This repository contains only Rebyte's management CLI and optional browser UI layers.

• Agent API documentation
• Getting started
• Create an API key

Two interfaces

Task	Interface
Execute an Agent	Official OpenAI SDK and Rebyte /v1/responses
Configure an Agent	Rebyte CLI, agent.toml, or Agent REST API
Add headless React chat state	@rebyte/agent-react
Add the optional chat UI	@rebyte/agent-ui

Execute with the OpenAI SDK

pnpm add openai

import OpenAI from 'openai'

const client = new OpenAI({
  apiKey: process.env.REBYTE_API_KEY,
  baseURL: 'https://api.rebyte.ai/v1',
})

const stream = await client.responses.create({
  model: process.env.REBYTE_AGENT_ID,
  input: 'Inspect this repository and summarize it.',
  stream: true,
})

let conversation
for await (const event of stream) {
  if (event.type === 'response.output_text.delta') {
    process.stdout.write(event.delta)
  }
  if (event.type === 'response.completed') {
    conversation = event.response.conversation?.id
  }
}

model is the Rebyte Agent ID. Pass the returned conversation on later turns. Rebyte supports a focused OpenAI Responses subset; Agent tools, Skills, MCP servers, and credentials come from the managed Agent configuration.

Configure with the Rebyte CLI

The CLI is a public, token-free GitHub Release asset.

pnpm add --global https://github.com/ReByteAI/rebyte-agent-toolkit/releases/latest/download/rebyte-cli.tgz
export REBYTE_API_KEY="rbk_..."
rebyte agent create -f agent.toml

The CLI validates, creates, applies, and exports organization-scoped API Agents. It does not execute Responses.

Headless React

pnpm add https://github.com/ReByteAI/rebyte-agent-toolkit/releases/latest/download/rebyte-agent-react.tgz

The browser talks to an endpoint in your application. That endpoint uses the official OpenAI SDK and keeps the organization API key on the server.

import { useMemo } from 'react'
import { createFetchTransport, useRebyteChat } from '@rebyte/agent-react'

function Chat() {
  const transport = useMemo(
    () => createFetchTransport({
      url: '/api/responses',
      interruptUrl: '/api/conversations/interrupt',
    }),
    [],
  )
  const chat = useRebyteChat({ transport })

  return (
    <form onSubmit={(event) => {
      event.preventDefault()
      void chat.send('Hello')
    }}>
      {chat.messages.map((message) => (
        <p key={message.id}>{message.content}</p>
      ))}
      <button disabled={chat.status === 'streaming'}>Send</button>
    </form>
  )
}

The React package contains browser state and SSE reduction only. It is not a second server-side Responses client.

Optional UI

pnpm add \
  https://github.com/ReByteAI/rebyte-agent-toolkit/releases/latest/download/rebyte-agent-react.tgz \
  https://github.com/ReByteAI/rebyte-agent-toolkit/releases/latest/download/rebyte-agent-ui.tgz

import { createFetchTransport } from '@rebyte/agent-react'
import { AgentChat } from '@rebyte/agent-ui'
import '@rebyte/agent-ui/styles.css'

const transport = createFetchTransport({
  url: '/api/responses',
  interruptUrl: '/api/conversations/interrupt',
})

export function App() {
  return <AgentChat transport={transport} agentName="Research Agent" inspector />
}

Run the example

Requires Node 22+ and pnpm 10.

pnpm install
pnpm build
cp examples/react-chat/.env.example examples/react-chat/.env.local
# Add REBYTE_API_KEY and REBYTE_AGENT_ID.
pnpm dev

Open http://127.0.0.1:4100. The server on port 4101 uses the official OpenAI SDK and forwards the SSE body without exposing the organization key.

pnpm test:live

See docs/architecture.md for the trust boundary.

Development

pnpm test
pnpm typecheck
pnpm build

MIT