How Was Amia AI Bot Setup?

,

Hello, this is off-topic and not related to trading, but at my job I’m trying to build a helpful AI agent bot and Amia inspired me. It (or she?) seems to respond to threads like a regular staff; creating tickets, offering suggestions and providing updates etc.

Can you give us insight on how it was built? Copilot Studio? ChatGPT agent? How does it have access to your knowledge base and support tickets? Any tips are appreciated.

Thanks.

For Amia we didn’t need to code anything, it’s all build on n8n.

Amia has two functions:

  1. Replies when someone post a ticket, checks that they provided the bot URL if they mention a bot in the ticket.
  2. Replies to the user when the dev resolves the ticket.

However it’s all outgoing communication, she will not respond if you text her (it’s possible but was too much work for our use case).

The way it works is:

  • For 1. there is a webhook triggered when someone posts in bug reports. This webhook is received by n8n, which then creates a new task in ClickUp. It uses the AI agent node to elaborate an acknowledgment reply, which is sent to Discourse and posted as Amia.
  • For 2. it’s triggered by the dev. When he marks a bug as “fixed”, he writes a short comment like “problem with bot service fixed now” and that is sent as a webhook from clickup to n8n, which gets beautified by another agent and posted as Amia.

Amia specifically doesn’t have a knowledge base hooked, but Max gain does (the AI inside Gainium dashboard). We have silently upgraded Max, he now have access to your account. Max was coded mainly with Copilot (Claude 4). The knowledge base was vectorized and saved in a qdrant database, I then created an API on top of qdrant so Max could query it.

If you share your specific use case I can give you more guidance based on my experience.

2 Likes

Thanks for the insight! Im not a coder myself, but do you think I could vibe-code my way through vectorizing company docs (word, pdf, excel etc) and building something for AI to query?

Yes you definitely can, I vibed coded everything. It just depends on how you want to set it up. For a chatbot with RAG and little setup I would use OpenAi’s assistants, you can create one in the playground and add docs. Another option is to use an external service as mcp for the chatbot. Both require minimal code, use Claude Sonner 4 to design the chatbot, just give your api keys. The chatbot in Gainium.io frontpage took me 30 minutes to set up.