When to Use Connector, Knowledge Base, or Flow
Your AI Agent answers far better when you give it the right material to work with. SIPESAN offers three things for that — the Knowledge Base, Connectors, and AI Flows — and it's easy to be unsure which one to reach for. The good news: these aren't an either/or choice. They complement each other. This guide helps you know when to use each, with examples that are easy to picture.
The big picture
Think of your AI Agent as a sharp new hire who, on day one, knows nothing about your business yet. These three things are what make it genuinely useful:
- Knowledge Base — what your business knows. The documents you write: product lists, prices, policies, how to order, and FAQs. The AI searches them by the meaning of the question, then answers from there. The content is fairly stable and the same for every customer.
- Connector — data that lives in another system. It links the AI to your API or systems (ERP, inventory, order system). Use it for data that keeps changing and needs to be checked case by case: order status, remaining stock, balance, or current pricing.
- AI Flow — a sequence you control. A fixed set of steps: ask this first, then check that, then reply like so. Use it when you don't want to leave the ordering up to the AI.
An easy way to remember it: Knowledge Base and Connectors are sources of material — where the AI gets its answers. An AI Flow is how you control the steps — how the conversation unfolds.
When to use the Knowledge Base
Use the Knowledge Base for things you can write down and that rarely change. This is knowledge, not numbers that move by the minute.
Good for:
- Explaining a product or service
- Prices and packages that stay fairly fixed
- Policies — warranty, returns, shipping
- How-tos and steps — how to register, how to pay
- Frequently asked questions (FAQ)
Example: a customer asks "what's the difference between plan A and plan B?" The AI searches the Knowledge Base and explains from your documents instead of guessing. The more complete your documents, the less often the AI has to say "sorry, I'm not sure".
See the AI Knowledge Base guide for details.
When to use a Connector
Use a Connector when the answer keeps changing, or has to be checked directly in another system using a code or number.
Good for:
- Order status or booking code
- Remaining stock or quota
- Balance, points, or invoices
- Prices that change often or in real time
- Any data that lives in your systems, not in the AI's "head"
Example: a customer writes "check order status ORD123" — the AI calls your order API with that code, then answers with the latest status. A simple rule of thumb: if a question needs a code or ID that's different every time, it's almost certainly a Connector (the HTTP API type), not the Knowledge Base.
See the AI Connector guide for details.
When to use an AI Flow
Most needs are already covered by the Knowledge Base plus Connectors — the AI shapes the conversation naturally on its own. You only really need a Flow when you want to lock down the order of steps.
Use a Flow when:
- There's a mandatory procedure that can't be skipped — for example, verify identity before showing any data.
- You need to collect several pieces of data in sequence, like a guided form.
- There's branching based on a result — if the status is "cancelled" offer A, if "shipped" offer B.
- You want the order and format of the answer to always be the same and consistent.
A Flow is triggered by a keyword in the customer's message, then runs the steps you designed — including calling a Connector through a fetch_data node.
A common mix-up: if you just want the AI to "check something when asked", that's not a job for a Flow. A Connector is enough. Wrapping it in a Flow only makes the conversation feel stiff.
See the AI Flow guide for details.
Example: one conversation, several sources
In practice, a single customer question often draws on more than one source at once.
Customer: "Is the 20GB internet plan still available at the Bandung outlet? How much is it?"
- The AI opens the Knowledge Base → it knows the 20GB plan's details and price from your documents.
- The AI calls a Connector → it checks live stock at the Bandung outlet from your inventory system.
- The AI combines both into one complete answer.
That's the strength of combining them: the Knowledge Base gives the stable context, the Connector gives the current numbers, and — when needed — a Flow keeps the steps tidy.
Quick summary
- Can be written down and rarely changes → Knowledge Base
- Keeps changing, or checked by a code in another system → Connector
- Needs a fixed order of steps → AI Flow
- Torn between a Connector and a Flow? If the AI can decide on its own when to fetch data, just use a Connector. Only when the steps must be tightly guided from start to finish should you wrap it in a Flow.
The three most common mistakes:
- Putting fast-changing data (stock, order status) in the Knowledge Base. The answers go stale quickly — that's a Connector's job.
- Building a Flow for something simple that only needs to check one piece of data. It makes the conversation stiff for no reason.
- Forgetting to fill the Knowledge Base. The AI ends up with no context and often says it doesn't know, when the answer was in your head all along — it just needed writing down.