Available on request
Call an agent computer from n8n, Make or Zapier
With the Sanbox n8n node, agent work becomes a workflow step: add the node, pick a template, pass the task and the files, and the next node receives the agent's results. The node is enabled per organisation on request. Today, a self-hosted n8n instance can run the Sanbox CLI from its Execute Command node, and Make or Zapier can hand work to that n8n workflow through a webhook.
What does the n8n node give you?
- Agent work as a workflow step. Send a task and input files to a Sanbox Computer, wait for it to finish, and pass the output files to the next node.
- Your templates and limits. The work runs from a template you choose, with its model, network policy and secret proxies.
- No shell access needed. The node works without enabling command execution on your n8n server.
- Safe retries. A repeated step returns the run it already started instead of starting a second one.
Enable it for your organisation
Enable it for your organisation and tell us whether you use n8n Cloud or self-hosted n8n, and what your workflows send to the agent. We'll set it up with you.
What works today: the Execute Command node
n8n's Execute Command node runs a shell command on the machine that runs n8n. Some things to know from n8n's docs:
- Self-hosted only. The node isn't available on n8n Cloud.
- Off by default since n8n 2.0. It's in the default
NODES_EXCLUDElist together with the Local File Trigger. To use it, setNODES_EXCLUDEto a list withoutn8n-nodes-base.executeCommand. n8n warns that the node can introduce significant security risks where users aren't fully trusted. - It runs inside the container. With n8n in Docker, commands run in the n8n container, not on the Docker host, so the Sanbox CLI has to be installed in a custom n8n image.
Then:
- Install the CLI in your n8n image with
npm install -g @sanlabs/sanbox-cli@latest. - Create an organisation API key in the Sanbox console and set it as
SANBOX_API_KEYin the n8n container's environment. Don't put the key in the workflow itself. - Create an OpenCode task template for the work, and note its ID from
sanbox templates list --json. - In the Execute Command node, run the task and wait for it:
cd /data && sanbox run "Summarise the files in /workspace/input and write output/summary.md" \
--template <template-id> --external-run-id <stable-id> \
--input report.pdf --wait --json
Run the command from the folder that holds the files, and pass --input paths relative to it.
With --json, the command prints one JSON document (with ok, data and next_actions fields) that the next node can parse. Use a stable --external-run-id, such as the ID of the record that started the workflow: if the step runs again, Sanbox returns the existing run instead of creating a duplicate.
- Download the results in a second Execute Command node:
sanbox runs download <run-id> --output /data/sanbox/<run-id> --json
From Make or Zapier
Make and Zapier are hosted services, so the simplest path today is to let them call the n8n workflow above. Start it with n8n's Webhook node, and call its URL from Make's HTTP "Make a request" module or from a Webhooks by Zapier action. Treat the webhook URL like a password.
Related
- Run an AI agent on a schedule
- Use Sanbox from any MCP client
- How do I run 10 coding-agent tasks in parallel without them clashing?
- Trigger an AI agent from Slack, Microsoft Teams or WhatsApp
Sources
- n8n Docs, Execute Command (self-hosted only, disabled by default from n8n 2.0, runs inside the n8n container, security warning), checked 2026-11-06
- n8n Docs, Nodes environment variables (
NODES_EXCLUDEdefault), checked 2026-11-06 - n8n Docs, Webhook node, checked 2026-11-06
- Make, HTTP app ("Make a request" module), checked 2026-11-06
- Zapier Help, Send webhooks in Zap workflows, checked 2026-11-06