Cloud agents on your own subscription

A cloud agent is a coding agent running in its own Tangle sandbox: a Linux machine with a filesystem, shell, and git that keeps working after your laptop closes. The agent can pay for its model with a plan you already have, so Tangle bills only the sandbox’s compute.

Credential	Harness	Model usage is billed to
Claude Pro or Max (`claude setup-token`)	Claude Code	Your Claude plan
ChatGPT plan (`codex login`) or OpenAI key	Codex	Your ChatGPT plan or OpenAI account
Z.ai GLM, Kimi, MiniMax, or any OpenAI-compatible key	OpenCode	That provider
No credential	OpenCode	Your Tangle credit, through the Tangle Router

Store the credential once

Credentials are stored as encrypted account secrets. A sandbox request names the secret; the credential never appears in the request.

# Claude Pro or Max
claude setup-token
tangle secret create CLAUDE_TOKEN --value-stdin <<< "$CLAUDE_CODE_OAUTH_TOKEN"
 
# ChatGPT plan
codex login
tangle secret create CODEX_AUTH --value-stdin < ~/.codex/auth.json

Start the agent

import { Sandbox } from "@tangle-network/sandbox";
 
const client = new Sandbox({
  apiKey: process.env.TANGLE_API_KEY!,
  baseUrl: "https://sandbox.tangle.tools",
});
 
const agent = await client.create({
  backend: {
    type: "claude-code", // or "codex" with format: "bundle" and CODEX_AUTH
    cliAuth: { account: "me", secretEnv: "CLAUDE_TOKEN", format: "token" },
  },
  secrets: ["CLAUDE_TOKEN"],
  git: { url: "https://github.com/your-org/your-repo" },
  idleTimeoutSeconds: 900,
});
await agent.waitFor("running");
 
const result = await agent.prompt("Find the failing test and fix it.", { sessionId: "main" });
console.log(result.response);

From a terminal, tangle sandbox create --backend claude-code --cli-auth CLAUDE_TOKEN starts the same agent, and tangle agent prompt <sandbox-id> "..." --session main --stream talks to it.

Coding plans

OpenCode reaches Z.ai’s GLM Coding Plan directly and any OpenAI-compatible endpoint through openai-compat. Store the key as a secret and name it with apiKeyEnv:

const agent = await client.create({
  backend: {
    type: "opencode",
    provider: "openai-compat",
    baseUrl: "https://api.moonshot.ai/v1", // Kimi; MiniMax: https://api.minimax.io/v1
    modelId: "kimi-k2-0905-preview",
    apiKeyEnv: "MOONSHOT_KEY",
  },
  secrets: ["MOONSHOT_KEY"],
});

Full guides

The Sandbox docs cover each path with streaming, profiles, egress rules, and spin-down:

Quickstart Overview