n8n

n8n is a workflow automation tool. With Nod, your n8n workflows can pause to ask humans for approval, send status updates to their phone, and wait for decisions — then continue based on the response. All through simple HTTP Request nodes calling the Nod API.

This template is a complete webhook agent with three event handlers: user messages, decision responses, and scheduled tasks. Download it and import into n8n to get started.

1. Create an agent in the Nod app. Note the agent ID and secret shown after creation.
2. Import the template into n8n (Settings → Import from File).
3. Set your OpenAI credential — n8n will prompt you to select or create one for the AI nodes.
4. Set your credentials in one place — open the Credentials node (the first node after the Webhook). Replace the placeholder values with your own:

   Field	Replace with
   agentId	Your Nod agent ID (e.g. agt_abc123)
   agentSecret	Your Nod agent secret (e.g. nod_abc123...)
   serperApiKey	Your Serper.dev API key

   All other nodes reference these values automatically — you only set them once.

5. Activate the workflow and copy the production webhook URL from the Webhook node (not the test URL).
6. Set the webhook URL on your agent in the Nod app (Agent Settings → Webhook URL).
7. Send a message from the Nod app — the workflow will process it and respond.

Every Nod event (messages, approvals, tasks) arrives at the same webhook URL as a POST request. The workflow routes each event to the right handler based on body.type.

Architecture

Nod Webhook
  → Credentials (sets agent ID, secret, API keys — one place for all)
  → Is User Message?      → [message handling pipeline]
  → Is Decision Resolved? → [approval handling pipeline]
  → Is Task Due?          → [task execution pipeline]

The three If nodes evaluate in parallel. Only one matches per event, so only one pipeline runs.

1. User messages

When a user sends a message from the Nod app:

1. Processing: Start — shows "Thinking..." in the app
2. Fetch Conversation History — gets the last 20 messages for context
3. Build Context — formats the history and latest message for the AI
4. AI Agent (OpenAI) — processes the message. If it needs current information, it responds with SEARCH: <query>
5. Needs Web Search? — checks if the AI response starts with SEARCH:

If no search needed: sends the AI reply directly, then processing_stop clears the indicator.

If search needed: sends an approval request to the user's phone, then processing_stop. The workflow ends here — when the user responds, a new decision_resolved event triggers the next pipeline.

2. Decision responses

When the user taps Approve or Reject on a pending request:

• Approved → a Research Agent runs with OpenAI and the Serper web search tool, then sends the result as a message
• Rejected → a fallback AI answers from its own knowledge (no search), then sends that instead

Both paths end with processing_stop to clear the indicator.

3. Scheduled tasks

When a scheduled task fires:

1. Task: Processing Start — shows the task is running
2. Task Agent — OpenAI with web search (no approval needed — tasks search freely)
3. Complete Task Run — calls POST /api/agent/tasks/runs/:run_id/complete with the agent's output as the summary
4. Task: Processing Stop — clears the indicator

API features used

This template is a starting point. Here are ideas for extending it:

Add more tools

• Connect databases, APIs, or file storage as additional AI tools
• Add a code interpreter tool for data analysis
• Integrate with Slack, email, or calendar for the final output

User preferences and memory

• Store user preferences in a database (language, timezone, response style)
• Fetch preferences at the start of each execution to personalize responses
• Auto-allow certain actions based on stored preferences instead of asking every time

Task enhancements

• Use the task context scratchpad to persist state across multi-step tasks that pause for user input
• Send requests during task execution to ask for user input mid-task (they appear in the Requests inbox with a "Task" badge)
• Add error handling: if the AI agent fails, mark the run as failed with a summary

Custom event routing

• Add a branch for task_trigger events from external webhooks
• Route different task types to different AI models or tool sets
• Use status_text on processing_start for custom indicator labels ("Searching the web...", "Analyzing data...")