Early access

One agent.Any harness.

Spinup is the stable API for agent runtime. Swap its harness, see its history, revoke it cleanly.

Usage

Call it from anywhere

The same agent, three ways in. Create it, run it on any harness, and pipe structured output back into your product.

spinup.ts

import { Spinup } from "@getspinup/sdk";

const spinup = new Spinup({
  apiKey: process.env.SPINUP_API_KEY,
});

const agent = await spinup.agents.create({
  name: "release-bot",
  harnesses: ["openclaw", "hermes"],
  skills: ["github", "slack"],
});

const run = await spinup.agents.exec(agent.id, {
  harness: "openclaw",
  input: "Review PR #431 and post the risks to Slack.",
});

console.log(run.output);

No lock-in

Swap the harness. Swap the model. Keep the agent.

One agent definition. Install packages, store project secrets, and switch between OpenClaw and Hermes today. The harness inside the environment is swappable. The model it talks to can change. The agent stays the part that lasts.

OpenClaw

Hermes

Claude CodeComing soon

OpenCodeComing soon

PIComing soon

CodexComing soon

Deep AgentsComing soon

NanoClawComing soon

Full environment

A real computer for every agent

Not a thin wrapper around one model call. Each agent runs on a full OS with a filesystem, browser, package managers, and tools like ffmpeg and Playwright, while the harness and model can change underneath.

Your Spinup Agent's own computer

Filesystem, ffmpeg, Playwright, Python. A real environment with swappable harnesses.

One endpoint, swappable harness and LLM

One endpoint per agent. Pick OpenClaw or Hermes today, swap to another harness when it lands. The model the harness talks to can change underneath.

Plug-in skills

Plug in PostHog, Slack, GitHub. Capabilities that follow your agent across runtimes.

Persistent state, recorded runs

Files and state survive between runs. Every run recorded against the Spinup Agent, with one durable history per identity.

Start here

Understand the runtime before you choose the harness

Start with the core pages on isolation, persistence, and secure execution. Then see how OpenClaw and Hermes run on the same model today, with Claude Code and other harnesses coming soon.

Runtime foundations

What is a Cloud Agent Runtime?→What is an Isolated Agent Environment?→Persistent Sandbox for AI Agents→Secure Code Execution for AI Agents→MicroVM vs Containers for AI Agents→

Harness guides

Runtime for OpenClaw→Runtime for Hermes→Runtime for Claude Code→Runtime for Codex CLI→Runtime for Deep Agents→

How it works

Real workflows. One stable API.

Each use case is a stable POST endpoint. Spinup runs it in a dedicated cloud environment and returns structured output your product can pipe into dashboards, CRMs, or automations.

Support resolution

Route inbound threads to an agent runtime and get back a resolution draft, classification, and escalation decision.

POST /agents/support-triage/run

{
  "account_ref": "acct_82f1",
  "include": ["transcript", "account_state"],
  "thread_id": "t_9a2e"
}

OpenClaw

Response

{
  "confidence": 0.94,
  "escalate": false,
  "resolution": "drafted"
}

See all use cases

Early access

Run agents that outgrew the sandbox.

Request access. Each Spinup Agent is a durable, workspace-owned agent with its own environment, its own history, its own kill switch, behind one stable API.