The Problem Nobody Talks About:
Every growing company has the same quiet problem.
Someone joins the team. They have questions — lots of them. What is the leave policy? How do I claim medical insurance? What tools do we use? Who do I contact for IT support?
And every single time, someone else has to stop what they are doing to answer.
It is not a big deal once. But when it happens every week, across every team, with every new hire — it adds up fast. Productive hours disappear. HR gets the same questions on repeat. Managers become unofficial FAQ pages.
What Is CB Agent?
CB Agent is our internal AI assistant. It lives inside our Slack workspace and can answer any question about our company — instantly.
Leave policies. Insurance claims. Office rules. Onboarding steps. Password security. IT tools. Project processes. You name it.
It does not search Google. It does not guess. It pulls answers directly from our actual company documents — our Notion pages, our Google Drive files, our internal policies and SOPs — and gives structured, accurate answers in seconds.
The best part? It feels completely natural. You just open Slack, type your question like you are messaging a colleague, and CB Agent replies.
Why We Built It This Way
There are a lot of ways to build an internal knowledge tool. We could have built a simple FAQ page. We could have used a basic search tool. But we wanted something smarter — something that actually understands questions, not just matches keywords.
That is where agentic AI comes in.
An AI agent does not just retrieve information. It reasons. It pulls from multiple sources, understands context, and assembles an answer the way a smart human would. Ask it a simple question — it gives a simple answer. Ask it something complex — it connects the dots across different documents and gives you a complete response. This is exactly what CB Agent does.
How We Built It This Way
Qdrant — The Brain (Vector Database)
Qdrant — The Brain (Vector Database)All our company documents — from Notion, Google Drive, and other sources — are stored in Qdrant, a vector database.
A vector database is different from a regular database. Instead of storing data as rows and columns, it stores the meaning of text. This means when you ask a question, it does not just look for matching words — it understands what you are asking and finds the most relevant information, even if the exact words are different.
Think of it like this: if you ask “how do I take a day off,” it finds the leave policy — even if the document says “annual leave application process” and not “take a day off.”
n8n — The Engine (AI Agent Workflow)
n8n — The Engine (AI Agent Workflow)n8n is where the intelligence lives. It is an open-source workflow automation tool — but in this case, we used it to build our AI agent.
The n8n workflow does several things:
- Receives the question from Slack
- Sends it to an embedding model to understand the meaning
- Searches Qdrant for the most relevant documents
- Passes the question and context to a large language model (LLM)
- Returns a clean, structured answer back to Slack
Notion + Google Drive — The Knowledge Base
Notion + Google Drive — The Knowledge BaseAll our company knowledge lives in Notion and Google Drive. HR policies, onboarding guides, admin SOPs, IT documentation, project processes — everything.
We pushed all of this into Qdrant using an embedding model. Every time we update a document, we can re-sync it so CB Agent always has the latest information.
Slack — The Interface
Slack — The InterfaceWe chose Slack as the interface because that is where our team already spends their day. No new app to learn, no new login, no friction. They just message CB Agent like they would message a colleague.
What Our Team Actually Asks It
Here are some real examples of questions CB Agent answers every day:
1.How do I apply for casual leave and what are the conditions?”
→ It explains the full process including the portal, notice period, and approval rules.
2.I came to work late today. What should I do?”
→ It pulls the attendance policy and explains the makeup process step by step.
3.How do I claim my OPD medical insurance?”
→ It gives the coverage details, eligible expenses, and the claim submission process.
4.Should I share my password with a colleague?”
→ Short, clear answer straight from our security policy. No.
5.I just joined CodersBucket. What are the most important things I need to know?”
→ This one is the most impressive. It pulls from the employee handbook, onboarding guide, Slack usage policy, IT tools guide, and daily hours policy — and assembles a complete structured answer from multiple documents at once.
The Real Impact — What Changed
We did not build this to impress anyone. We built it because we had a real problem. Here is what actually changed after CB Agent went live:
1. Nobody has to stop their work to answer the same question twice
HR used to get pinged multiple times a day for the same policy questions. That stopped almost immediately. The team started asking CB Agent first — and getting answers faster than they would have by waiting for a human reply.
2. New joiners feel confident from day one
New team members often feel shy asking basic questions — they do not want to look like they do not know things. CB Agent removes that friction completely. They can ask anything, anytime, privately in Slack, and get a proper answer without judgment.
3. HR and admin focus on work that actually needs them
When HR is not spending time answering “how many leave days do I have,” they are free to focus on things that actually require a human — performance conversations, team culture, hiring. That is a real shift in how time gets used.
What Is Agentic AI and Why Does It Matter for Business?
You might have heard the term agentic AI a lot recently. Here is what it actually means in plain language.
Traditional AI tools answer questions based on what they were trained on. Agentic AI goes further — it takes actions, retrieves information from external sources, reasons across multiple steps, and delivers results that go beyond a single response.
CB Agent is a simple example of agentic AI in action. It does not just reply with a pre-written answer. It:
1. Understands the intent behind the question
2. Searches a live knowledge base for relevant documents
3. Synthesizes information from multiple sources
4. Delivers a structured, contextual answer
As AI technology moves forward, agentic systems will become the standard for how businesses operate internally — not just for answering questions, but for handling workflows, processing requests, and supporting decisions.
We are at the beginning of that shift. CB Agent is our first step.
How You Can Build Something Similar
If you are a founder, CTO, or team lead reading this and thinking “we need something like this” — the good news is that the stack we used is accessible, open-source friendly, and does not require a massive engineering team.
Here is the basic path:
1. Gather your documents — Notion, Google Drive, Confluence, PDFs, whatever you use
2. Choose a vector database — Qdrant is what we use; Pinecone and Weaviate are alternatives
3. Pick an embedding model — OpenAI embeddings work well, there are open-source options too
4. Build your agent in n8n — connect the pieces: Slack input → embedding → vector search → LLM → Slack output
5. Upload your knowledge base — chunk your documents, embed them, push to Qdrant
6. Test it with real questions — iterate until the answers are accurate and clean
The hard part is not the technology. The hard part is organizing your knowledge base well enough that the AI can find the right answers. If your documents are messy or outdated — the agent will reflect that.
Final Thought
CB Agent is not a product we are selling. It is a tool we built for ourselves — because we had a real problem and we had the skills to fix it.
But the bigger idea behind it matters. Every company has knowledge trapped in documents that nobody reads, policies nobody can find, and questions that eat up hours of productive time every week.
AI agents — built on tools like n8n, Qdrant, and connected to your existing knowledge base — can change that. Not in a distant future. Right now, with tools that exist today.
We are sharing this because we think more companies should be building this way. Start small. Solve a real internal problem. See what changes.
Built at CodersBucket using n8n, Qdrant, Notion, Google Drive, and Slack. Interested in building something similar? Reach out to our team.
