How can I build a multi-agent YogaAssistant in n8n to provide source URL recommendations using Supabase?

Below is one approach to design your YogaAssistant using n8n together with Supabase and Telegram. Remember that you can always adjust things as needed, but here are some recommendations:

---

1. Storing Links in Supabase

Instead of hardcoding the URLs inside your workflow, a good practice is to create a dedicated table in Supabase with the following columns:

• id – Primary key
• topic – The health problem or course theme (e.g., “Breathing Problems”)
• link – The URL to the dedicated page
• description – A brief summary of the page content

This way, you can easily update, add, or change recommendations without modifying your workflow. In n8n you can use either the Postgres node (if you’re connecting directly to your Supabase database) or the HTTP Request node (using Supabase REST API) to query this table.

Documentation Links:

• Supabase Getting Started
• Supabase Database Overview

---

2. Architecting Your Multi-Agent Setup

To implement a consistent multi-agent approach, split your process into three logical parts or “agents.” Each can be a separate workflow or a distinct section in one workflow:

a. Greeting Agent

• Responsibility: Collects the user's metadata (name, contact info) when they initiate a conversation.
• Nodes:
  • Telegram Trigger: To capture incoming messages.
  • Set/Function Nodes: To structure and store the user data.
  • Supabase/Postgres Node: Inserts the data into your users table.

b. Anamnes Agent

• Responsibility: Collects details about the user's health concerns, asking clarifying questions as needed.
• Nodes:
  • Telegram Node (for follow-up questions) or Switch/If Nodes: To evaluate responses and branch the conversation.
  • Function Nodes: To handle any data transformation or decision-making logic.
  • Data Store or Supabase Node: To record the anamnesis for later use if necessary.

c. Recommendation Agent

• Responsibility: Uses the stored information plus a Supabase query to pick 2–3 relevant course links.
• Nodes:
  • Supabase/Postgres Node or HTTP Request Node: Query your links table by matching topics with recognized keywords in the user’s anamnesis.
  • Merge Node: (If using separate workflows) Combine query results with user data.
  • Telegram Node: Send back the recommendation message containing the user's condition summary and the recommended links.

Tips:

• Use Trigger or Webhook nodes to transition between agents if you break your solution into several workflows.
• Branching logic (via If or Switch nodes) will be essential to ask clarifying questions when the user message is ambiguous.

---

3. Suggested Sequence of Nodes in Your Workflow

Below is a simplified sequence that you might employ (you can break these into separate workflows if preferred):

1. Telegram Trigger:
   Receive the initial message from the user.

2. Greeting Agent Section:

   • Set Node: Parse initial text to extract user metadata.
   • Supabase/Postgres Insert Node: Save user data in your Supabase database (users table).

3. Anamnes Agent Section:

   • Telegram Node (or additional Trigger): Ask for details about the user's symptoms.
   • If/Switch Node: Evaluate if clarifying questions are needed.
   • Function Node: Aggregate and format “anamnesis” data.

4. Recommendation Agent Section:

   • Supabase/Postgres or HTTP Request Node:
     Query your “links” table based on keywords detected in the anamnesis (e.g., if the user's message mentions “hip pain,” filter for courses on hip problems).
   • Merge Node: Combine the collected anamnesis data with the fetched links if needed.

5. Response Node:

   • Telegram Node: Send a composed message that includes a condition summary, recovery recommendations, and embeds the 2–3 generated links (each with its description).

6. Optional Logging/Audit:

   • Discord or Email Node: Notify your team or log the event for debugging purposes.

Example Pseudocode for a Function Node (selecting links):

JAVASCRIPTCopied
// An example snippet for selecting links based on keywords
const userMessage = $json["userAnamnesis"];
let selectedLinks = [];

if (userMessage.includes("hip")) {
  selectedLinks.push({ topic: "Hip Problems", link: "https://yoursite.com/hip-courses", description: "Courses focusing on hip pain and mobility" });
}
if (userMessage.includes("breath")) {
  selectedLinks.push({ topic: "Breathing Problems", link: "https://yoursite.com/breath-courses", description: "Courses dedicated to improving breathing techniques" });
}

return { selectedLinks };

You can also do this by directly querying your Supabase table instead of hardcoding logic here.

---

Final Thoughts

This architecture ensures that:
• Your course pages (links) are maintained in one place (Supabase), making updates easier.
• Each agent has a clear responsibility.
• The sequence of nodes makes it easy to track the flow of data from the initial Telegram message through to the final recommendations.

For more detailed instructions on node configuration, take a look at the n8n core nodes documentation and the Supabase docs linked above.

I hope this helps you design a robust solution for YogaAssistant. Good luck with your presentation!