Run OpenCode in a sandbox

OpenCode (opencode) is installed in the ubuntu template. Choose how it signs in, then run a task with opencode run.

Sign inCopy Anchor Link

With the API key of a providerCopy Anchor Link

OpenCode reads the key of each provider from its usual variable, such as ANTHROPIC_API_KEY or OPENAI_API_KEY. Put it in the env of the sandbox:

{
  "ANTHROPIC_API_KEY": "sk-ant-..."
}

opencode models lists the models that your keys reach. -m provider/model picks one for a run.

With a subscriptionCopy Anchor Link

Sign in inside the sandbox, over SSH:

• GitHub Copilot (Pro, Pro+, Business or Enterprise): run opencode auth login, choose GitHub Copilot, and enter the code that it prints at github.com/login/device.
• OpenCode Zen or OpenCode Go: create a key at opencode.ai/auth, and put it in the env of the sandbox as OPENCODE_API_KEY.

OpenCode keeps a sign-in in ~/.local/share/opencode/auth.json, and a persistent sandbox keeps it across pauses. A Claude subscription does not work in OpenCode, because Anthropic does not allow it. For a ChatGPT plan, use Codex. See the providers in the OpenCode docs.

With the open models of Munito AICopy Anchor Link

Put your Munito AI key in the env of the sandbox as MUNITO_API_KEY, and write this configuration to ~/.config/opencode/opencode.json in the sandbox, with the files API or a command:

{
  "$schema": "https://opencode.ai/config.json",
  "model": "munito/zai-org/glm-5.3-flash",
  "provider": {
    "munito": {
      "npm": "@ai-sdk/openai-compatible",
      "name": "Munito",
      "options": {
        "baseURL": "https://api.munito.ai/inference/v1",
        "apiKey": "{env:MUNITO_API_KEY}"
      },
      "models": {
        "zai-org/glm-5.3-flash": {
          "name": "GLM-5.3-Flash",
          "limit": {
            "context": 1048576,
            "output": 943718
          }
        }
      }
    }
  }
}

{env:MUNITO_API_KEY} reads the key from the environment, so the file holds no key. OpenCode with open models explains the provider, the reasoning effort and the regional hosts.

Run a taskCopy Anchor Link

curl -sS --fail-with-body "https://api.munito.ai/sandboxes/v1/sandboxes/$SANDBOX/exec" \
  -H "Authorization: Bearer $MUNITO_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"command": "cd /workspace && opencode run --auto \"Write a Python script that prints the first 10 primes, and run it\"", "timeout": 1800}'

• opencode run runs one task with the build agent and exits. It writes the answer to standard output.
• --auto approves each permission that no rule denies, so OpenCode runs commands and changes files without asking. Without it, opencode run rejects each permission that would ask.
• -m provider/model picks the model for one run. Without it, OpenCode uses model from its configuration.
• A command runs for at most 30 minutes. For a longer task, run opencode run in the background with nohup and read its log with later commands.

Use the terminal UI over SSHCopy Anchor Link

Add your public key with SSH, connect with the command that GET /sandboxes/{sandbox_id}/ssh answers, and run opencode in /workspace. For a session that outlives your connection, install tmux (apt-get install -y tmux) and start OpenCode inside it.