Run n8n, Hermes Agent, OpenClaw and Steel Browser on your own isolated infrastructure — created from the console in one click. Machine-readable copy at /docs/ai-apps.md.
| App | What it is | Where it runs | Credentials |
|---|---|---|---|
| n8n | Workflow automation (Zapier-style) with AI nodes, a visual editor and webhooks | Shares your account’s Docker environment — ready in seconds | None required |
| Hermes Agent | Autonomous AI agent (Nous Research, MIT) with persistent memory and skills; connects to Telegram, Discord, Slack, WhatsApp and email | Dedicated, isolated VM (~3–5 min) | Dashboard password + API server key (generated), plus at least one AI provider key |
| OpenClaw | Self-hosted personal assistant that runs scheduled tasks and replies over WhatsApp, Telegram, Discord and more | Dedicated, isolated VM (~3–5 min) | Access token (generated), plus an AI provider key |
| Steel Browser | Headless browser for AI agents — sessions API + CDP for Playwright/Puppeteer. Apache 2.0 | Dedicated, isolated VM (~3–5 min) | None required |
Every app gets a public URL under apps.lookerhost.com with HTTPS included, visible on its project card.
* are required.ANTHROPIC_API_KEY, OPENAI_API_KEY, OPENROUTER_API_KEY) are your own; get one from your provider. Agent apps require at least one.Your credentials are encrypted into the deployment and are never stored by the panel, never shown again, and never written to logs.
n8n — the first visit asks you to create the owner account. Webhook URLs are already generated with your public HTTPS domain, so triggers from external services work out of the box.
Hermes Agent — sign in to the dashboard with user admin and the password you saved from the wizard. From there, connect messaging channels (Telegram, Discord, Slack…) and chat with your agent. It also exposes an OpenAI-compatible API protected by the API server key you saved.
OpenClaw — the Control UI opens from your app URL; paste the access token you saved in the wizard. Connect channels (e.g. a Telegram bot token) and define scheduled tasks from there.
Steel Browser — unlike the other apps, Steel is not used by hand: it is driven by code. It is an isolated Chrome that your program (or your AI agent, or an n8n flow) commands over an API, and /ui is the live viewer where you watch what the robot is doing. Three ways to use it, easiest first:
# Extract a page's content (ideal to feed an AI)
curl -X POST https://YOUR-STEEL.apps.lookerhost.com/v1/scrape \
-H "Content-Type: application/json" -d '{"url":"https://example.com"}'
# Screenshot of any page
curl -X POST https://YOUR-STEEL.apps.lookerhost.com/v1/screenshot \
-H "Content-Type: application/json" -d '{"url":"https://example.com"}' -o shot.jpg
# Page → PDF
curl -X POST https://YOUR-STEEL.apps.lookerhost.com/v1/pdf \
-H "Content-Type: application/json" -d '{"url":"https://example.com"}' -o page.pdf
Full-control sessions — your code creates a session (POST /v1/sessions) and connects Playwright/Puppeteer over CDP: clicks, forms, human-like navigation. While it runs, open /ui to watch the robot’s screen live.
From another agent — an n8n or Hermes on the same account uses these endpoints as its “eyes on the web”: e.g. an n8n flow that scrapes competitor prices with /v1/scrape every morning and posts them to Telegram.
Agent apps normally need your own provider key (Anthropic, OpenAI, OpenRouter). If your account has AI credits, some apps can use those instead: pick Use LookerHost credits in the wizard and set a budget — the platform issues a scoped key against its own LLM gateway and meters usage from that budget, no key to manage.
Steel Browser works this way today. Hermes Agent does not: its own startup process ignores the credits key for its main agent and always talks to OpenRouter directly, so the option is not offered for it in the wizard — bring your own Anthropic, OpenAI or OpenRouter key for Hermes. OpenClaw follows its own SDK’s standard handling of a custom endpoint, so credits are expected to work there, though we have not run an end-to-end check of that specific path.
The same operation is available to scripts and agents. REST:
curl -X POST https://lookerhost.com/api/services/aiapps \
-H "Authorization: Bearer <API_KEY_WITH_apps:write>" \
-H "Content-Type: application/json" \
-d '{"label":"my-agent","app":"hermes","envs":{"HERMES_DASHBOARD_BASIC_AUTH_PASSWORD":"…","API_SERVER_KEY":"…","ANTHROPIC_API_KEY":"sk-ant-…"}}'
Via MCP, use the lookerhost_create_aiapp tool — the per-app credential spec comes back in aiApps from lookerhost_list_plans. See the MCP server and API documentation for keys, scopes and the full tool catalog. This means an agent you run on LookerHost can provision more LookerHost infrastructure for you.
Once a Hermes or Steel app is running, the same MCP server can talk to it directly — no need to expose its API to the internet yourself:
lookerhost_agent_chat — send a message to your Hermes Agent and get its reply.lookerhost_browser_session — create, list, get or release a Steel Browser session (returns a CDP endpoint for Playwright/Puppeteer).lookerhost_browser_scrape — render and extract a page with your Steel Browser in one call.lookerhost_aiapp_api — a lower-level call to any allow-listed route on either app’s API.These reach the app over LookerHost’s internal network; the agent’s own API is never exposed publicly.
See current AI app plans on the pricing page. Each plan covers the app, its (dedicated) compute, the public HTTPS URL and platform monitoring; AI usage is billed by your own provider through your key.