Hey folks, martech_ops_mike here. I've been lurking and seeing a lot of questions about getting started with OpenClaw, especially the Claw-Core model. It can feel overwhelming, so I wanted to share my first *working* setup. This isn't fancy, but it's a reproducible recipe for a simple ticketing bot that categorizes and routes internal support questions.
My goal was to stop manually sorting emails from our "help@" inbox into Notion. The bot needed to: 1) read a support request, 2) categorize it (e.g., "Billing," "Technical," "General"), and 3) push a structured ticket to a Notion database.
**My Toolstack & Configuration**
* **LLM:** OpenClaw Claw-Core (via API)
* **Trigger:** New email in a dedicated Gmail folder (via Zapier)
* **Orchestrator:** A simple Python script (could be Make or n8n too)
* **Destination:** Notion database
The magic is in the prompt/instruction for Claw-Core. After some testing, this structure worked reliably:
```
You are a ticketing classifier. Analyze the user's support request and extract the following:
- **Primary Category:** Choose ONLY one from [Billing, Technical Issue, Account Access, General Inquiry].
- **Urgency:** "High," "Medium," or "Low." High is for system outages or login failures. Low is for feature questions.
- **Summary:** A single-line, clear summary of the issue.
Format the output as valid JSON only, using the keys "category", "urgency", and "summary".
Do not include any other text.
```
My script takes the email body, sends it with this prompt to the Claw-Core API, parses the JSON response, and then uses the Notion API to create a new page with those properties. The category becomes a Notion select property, urgency a status, etc.
**Rationale & Results**
I started with a more complex prompt asking for multiple categories and a sentiment score, but Claw-Core would sometimes get "creative" and invent categories. Locking it down to a strict list and demanding pure JSON made the output 100% parseable.
* **Before:** 10-15 minutes a day sorting emails.
* **After:** Tickets auto-populate in Notion, already categorized. I only step in if the "urgency" flag is "High," which has been accurate about 90% of the time.
It's not perfect, but for a free, open-source model, it's a fantastic start. This basic structure could be adapted for CRM lead scoring or survey response sorting too.
What simple automations are you all building with the core models? Any tweaks you'd make to this prompt?
stay automated
stay automated