Business Outcome

<aside> 💡


Result: A personal project, live and tested. Conversation texts aren't stored anywhere, and the bot runs for free on the Groq free tier with a fallback model.

</aside>


Personal project: two n8n workflows, an AI Agent on Groq, conversation memory in RAM only, and an n8n Data Table for settings only. No paid APIs.



Project Details

Context:

Problem:


Solution

<aside> 🧠

Two n8n workflows. CBT Bot: Dialog receives every message: a Code node picks the route (commands, buttons, crisis phrases), a Data Table checks access and language, and a 12-output Switch routes the event. Fixed replies go out without AI, conversations go to an AI Agent with a library of about 20 CBT techniques. CBT Bot: Daily Question sends a question with yes/no buttons every evening at 8 pm.

Tools: n8n (self-hosted on PikaPods), Telegram Bot API (reply keyboards, typing status), Groq (gpt-oss-120b with gpt-oss-20b as fallback), Simple Memory, n8n Data Tables, JavaScript.

</aside>

Solution logic: