Durable Actors

an open-source alternative to Cloudflare Durable Objects... with no vendor lock-in, memory limits, and observability built in.

Durable Actors help you build real-time applications like chat systems (e.g. ChatGPT, Codex), collaboration tools (e.g. Notion), and agent swarms (e.g. Devin).

They provide stateful serverless functions, a foundational building block that abstracts away persistence, coordination, and infrastructure challenges in distributed systems.

Live demo · TypeScript documentation · Python documentation

Hosted version

Try Durable Actors on Terse Cloud at app.useterse.ai. To get started, run:

bunx terse-cli@latest install

How it works

1. Define an actor, a class with durable state (i.e. data survives interruptions, errors, and restarts) and serialized execution (i.e. concurrent callers can update it safely). For example:

• If you were building ChatGPT... a chat actor can store conversations that survive LLM flakiness and server crashes (durable state)
• If you were building Notion... a document actor can coordinate concurrent edits from several people and agents (serialized execution)

2. Generate type-safe clients automatically with the Durable Actors SDK. For now, it supports Python and TypeScript.
3. Develop locally with one command and later self-host the Durable Actors runtime on GCP.

Quickstart: Multiplayer AI Chat

1. Create your project

Install Bun 1.4.2+.

bunx durable-actors init my-actors
cd my-actors
bun install
bunx durable-actors dev # Run the server locally on your machine

2. Define an actor

Define and export actors in your actor project’s src/actors.ts, the default entrypoint loaded by durable-actors dev. The runtime loads actors on demand and persists fields marked @Persisted.

For example, a chat history actor:

import { openai } from "@ai-sdk/openai"
import { streamText } from "ai"
import { Actor, type ActorSocket, Interleave, Persisted } from "durable-actors"

type Message = { role: "user" | "assistant"; content: string }

export class ChatHistory extends Actor<null, string, Message[]> {
    @Persisted messages: Message[] = []

    async onConnect(socket: ActorSocket<null, Message[]>) {
        socket.send(this.messages)
    }

    @Interleave // Let other calls run during await, e.g. new connections while a reply streams.
    async onMessage(_socket: ActorSocket<null, Message[]>, text: string) {
        this.messages.push({ role: "user", content: text })
        const result = streamText({ model: openai("gpt-5-mini"), messages: [...this.messages] })
        const reply: Message = { role: "assistant", content: "" }
        this.messages.push(reply)
        this.broadcast(this.messages)
        for await (const chunk of result.textStream) {
            reply.content += chunk
            this.broadcast(this.messages)
        }
    }
}

3. Connect your backend

We make it super easy to integrate the actors into your existing tech stack. Just generate the client and you get a fully type safe contract to interact with.

bunx durable-actors generate

Create a WebSocket connection to one shared chat:

import express from "express"

import { actors } from "../generated/index.js"

export const app = express()

app.post("/api/chat/socket", async (_req, res) => {
    const grant = await actors.ChatHistory.prepareWebsocket({
        actorId: "lobby",
        metadata: null
    })
    res.set("Cache-Control", "no-store").json(grant)
})

4. Connect the frontend

import { useEffect, useRef, useState } from "react"
import { createRoot } from "react-dom/client"

import type { actors } from "../generated/index.js"

function Chat() {
    const socket = useRef<WebSocket>(null)
    const [messages, setMessages] = useState<actors.ChatHistory.Outgoing>([])

    useEffect(() => {
        let active = true
        async function connect() {
            const response = await fetch("/api/chat/socket", { method: "POST" })
            const { websocketUrl } = await response.json()
            if (!active) return
            socket.current = new WebSocket(websocketUrl)
            socket.current.onmessage = event => setMessages(JSON.parse(event.data))
        }
        void connect()
        return () => {
            active = false
            socket.current?.close()
        }
    }, [])

    function send(form: FormData) {
        const text = String(form.get("message")).trim()
        if (!text || socket.current?.readyState !== WebSocket.OPEN) return
        socket.current.send(JSON.stringify(text))
    }

    return (
        <main>
            <div role="log" aria-label="Messages">
                {messages.map((message, index) => (
                    <p key={index}><strong>{message.role}:</strong> {message.content}</p>
                ))}
            </div>
            <form action={send}>
                <input name="message" aria-label="Message" required />
                <button>Send</button>
            </form>
        </main>
    )
}

createRoot(document.getElementById("root")!).render(<Chat />)

For a complete app with error handling and retries, see the AI Chat example.

5. Monitor and debug

Durable Agents provides built-in observability features:

Examples

For complete sample applications, see AI Chat, Collaborative Documents, and Chatroom.

Build with a coding agent

Projects created with durable-actors init already include the skill. To use it elsewhere in Claude Code, install the plugin:

/plugin install durable-actors --marketplace TerseAI/durable-actors

For Cursor, Codex, and other agents, install the skill with skills.sh:

npx skills add TerseAI/durable-actors

Then describe what you want to build, or run /durable-actors:durable-actors-quickstart in Claude Code.

Community

Bug reports, feature requests, documentation fixes, and code contributions are welcome. See the contributing guide for repository setup and checks, and follow our code of conduct.

Use GitHub Issues for bugs, ideas, and questions. Report vulnerabilities privately using our security policy.

Follow development and release notes on GitHub Releases.

License

MIT © 2026 Terse