Hey folks,
I’m exploring ways to create a smart knowledge base that helps automate customer support for SaaS apps.
Curious how you handle it:
Do you store info in docs, code, or a database?
Any AI or automation tools you’ve tried?
How do you keep everything up to date?
Would love to learn from your workflows and ideas.
I dont have a huge volume of support requests - product was designed to avoid the need for support. But I did build out a help center to defer contacts and to give me a fast way to link customers to answers. I’ve run a bunch of larger businesses with big support and cs orgs so know what I’m trying to avoid (costs and tools and people). I put faqs in a python file for fast type ahead search and made my own tool and chrome extension for capturing step by step instructions. Here’s where it lives - covers the topics that most frequently came up from users and I add an faq every week or so if new advanced features are added. https://grasshoppersignup.com/faq
Knowledge-backed AI agents work well for this. You feed them your docs/code/FAQ data and they handle common support questions automatically, cutting response times way down.
The tricky part is keeping the knowledge current without constantly retraining. Look for systems where you can update the knowledge base and it propagates immediately - no model retraining cycles. Weavy’s AI Copilot component does this with OpenAI/Claude/Gemini backends, and you can embed it right in your SaaS interface so users get help without leaving your app.
What’s your current support volume looking like?
For large applications/services I’ve found ChatGPT is way more accurate than the company’s own documentation.
If it’s your own app and it’s too small to be recognized on ChatGPT, I’ve looked at Scribe in the past. I think there are a few tools out there where you can do a workflow on desktop and it’ll automatically put documentation together based on your desktop actions (Microsoft even has one I believe).
If I was being fancy, I’d build a RAG chatbot into my app. Not difficult at all in the grand scheme of things.
That’s a really smart approach — designing the product to minimize support needs is the best kind of automation.
I love how you built your own lightweight FAQ system in Python — that’s super efficient compared to using heavy third-party tools. The Chrome extension idea for capturing steps is clever too!
Do you also connect these FAQs to any AI assistant or plan to make it searchable via chat? I’m exploring something similar with Vezlo, which turns FAQs and docs into conversational responses.
Thanks, Rickardh — totally agree on the retraining issue.
That’s actually one of the problems I’m exploring — how to sync real-time updates from docs or databases without full retraining cycles.
I haven’t tried Weavy’s Copilot yet, but it sounds like a solid embedded option.
Right now, support volume is still moderate — under a few dozen tickets a week — but I’m aiming to automate more before scaling.
Appreciate the insight!
That’s a great point, Shawn! 
Yeah, ChatGPT does surprisingly well with well-known products — it’s like it already has the company wiki built in. I’ll definitely check out Scribe, sounds like a smart way to auto-generate docs from workflows.
I’m actually thinking about connecting these kinds of auto-generated guides into a searchable knowledge base for customer support — maybe even sync it with an AI assistant. Have you tried anything like that before?
That’s a great idea!
I’ve been exploring RAG-based setups too — especially how to connect the data layer efficiently.
Curious though, what stack or tools would you recommend for embedding and retrieval? (Vector DB, framework, etc.)
My pleasure!
I am in the middle of building one in Replit, using:
claude sonnet 4.5 for the LLM
supbase for vector storage
voyageai.com for doing the embedding
The admin page of the app let’s the app owner re-upload their 20 markdown training files whenever they require, and kick-off a new build/embed. I know some people would need this to be automated, but my client doesn’t intend to updated these docs very often, so a manual process made more sense.
I have been debating whether to make them more accessible directly in the form builder - but I end up (so far) deciding that if someone needs a help center to build a form, its a UX problem that needs to be solved in a more clever or obvious way. I do like the concept of using chat to get quick answers rather than pull them out of the main experience. For me its a balance of wanting them to learn and explore vs. just getting them their answer quickly. I’ll check out Vezlo too
That’s a solid setup! Using VoyageAI with Supabase makes the pipeline really clean. I’m building something similar with Vezlo, where the AI Assistant Server handles the embeddings, chat history, and vector search — trying to make setups like yours easier to deploy.
That’s a great.
Would love to hear your thoughts once you check it out!
Just been reading your website - a lovely page. But I am a bit confused who your market is to use this?
Surely if I want to learn about a new codebase I am working on as an engineer I just ask the AI agent to analyse it for me and tell me what I need to know. Why would I need another level of extraction.
Clearly I’ve missed something😊
You’re right—AIs can read code, but Vezlo goes further. It builds a shared knowledge base, connects with an assistant server for live Q&A, and keeps everything open source.
cool. but for who?
Mainly for SaaS dev teams who want their code, docs, and support content in sync. Vezlo helps them power in-app help or chat without managing separate systems.
Maybe integrate it through n8n with Google Drive, so it automatically imports a folder, sends the content into Pinecone, and connects it to a chatbot as a vectorised knowledge base. You could also set it to check for updates in Google Drive nightly, syncing new or changed files to keep the client’s data current. I have 3 n8n workflows for this if you want them.
That’s a solid setup. I’ve used n8n for smaller automations but not full syncs like this. Would love to check out your workflows — could help me streamline how Vezlo handles content updates too.