AI voice bot database lookup: how to connect and test it

AI voice bot database lookup is what you need when there is so much data that the bot starts mixing it up. A real estate agency has hundreds of apartments, a taxi fare depends on two addresses, a clinic or salon runs the schedules of a dozen specialists, and a store has a catalog of thousands of products. You can't load all of that into the bot at once: the call gets pricier, the bot confuses dates and prices, and the client goes to a competitor.

With a dynamic database, the bot finds the two or three rows it needs right during the conversation — apartments within budget, the fare, a stylist's free time, stock on hand — and books the client for an exact time by itself. This is the voice bot's main advantage: large live databases with no made-up answers. There is no separate fee — only conversation minutes, $0.12 per minute on average. Watch the two-minute video, and below you'll find the developer brief, the setup and a breakdown of mistakes.

What a dynamic database is and how it works during a call

A dynamic database is a search in your data during the call through a webhook: a link the platform uses to query your server by itself. On the bot card, it is a trigger with the database search action. Here is how it works:

1. A keyword is said — by the client (“any free time tomorrow?”) or by the bot (“Checking the schedule...”).
2. AI uses the parameter prompts to take the date, service, district or item number from the conversation.
3. The platform sends a GET request, and the bot says a short pause phrase like “one second, let me check”.
4. Your server returns a short answer: three free slots, five apartments, the stock of an item.
5. The bot answers from the “Dynamic webhook instruction” — only from this data, so it doesn't mix up dates.

When you don't need a dynamic database

First check whether a regular loading webhook is enough — it usually is. Connect a Google document with your data (“Webhooks” → the loading webhooks block) and ask the bot about dates, services and prices in “Training AI”. If it doesn't get confused, you don't need a developer. The platform recommends keeping loading data within 50 KB and 7 seconds, and it won't save more than 100 KB or 10 seconds. Read more in “AI voice bot knowledge base”.

For Yclients you don't need a developer either: the wizard on the “Integrations” tab creates the database search trigger by itself — a ready-made example of a dynamic database, see “Yclients integration”. Free time from Google Calendar comes in through its own wizard: “Google Calendar integration”.

Developer brief: three endpoints and the response format

A developer writes the middleware — a small piece of code between the bot and your database. The database can be anything: a spreadsheet, XML, MySQL, a CRM or a booking service with an API. First list the search parameters (date, service, staff member), then hand over the task. Booking needs three endpoints:

• A directory of services and staff with their IDs goes into a loading webhook: AI uses it to turn the client's words into an ID.
• A slot search, for example https://example.com/slots?date=24.03.2026&service_id=3, goes into the database search trigger.
• The booking itself goes into “Webhooks” → the write webhook block.

Requirements for the search response:

• Short: up to three slots closest to the requested time; for a catalog, 3–5 items.
• Plain text, JSON or XML with clear field names; in our setup experience, the bot reads XML most reliably.
• Instead of “error” or code 500 — a phrase with a hint, so the bot asks the client again by itself.
• Fast and GET only: the client is waiting on the line the whole time.

The trigger's ready-made answer instruction understands the freeSlots block. Before setup, open every link in a browser: remove the date, change the service, pick a day off. You can build the middleware yourselves or order a turnkey setup with the button at the end of the article.

How to set up the database search trigger

1. 1
   Add a trigger

   “Triggers” tab → “Add trigger”. A new trigger gets the database search action and a demo database right away: the bot asks for an account number and reads out the balance.
2. 2
   Enter the keywords

   Fill in the “Trigger name”, for example “Schedule”, and the trigger keywords field: the phrase the bot says before the search (“checking the schedule”) and, if you like, client phrases.
3. 3
   Check the pause

   The pause filler phrase: a new trigger already has a short “one second, let me check” phrase with a 2,000 ms delay.
4. 4
   Paste the link

   The search address goes into the dynamic database request URL field, method GET. Put constant values straight into the link: city=nyc.
5. 5
   Add parameters

   Delete the demo parameter akk, click “Add parameter”, and set the parameter name in Latin characters and the parameter prompt. The parameter is added to the link by itself: date=${date}.
6. 6
   Describe the answer

   “Dynamic webhook instruction”. If left empty, a ready-made instruction for booking slots is used; for a catalog or stock, rewrite it. Click “Save”.

An enabled trigger without keywords won't save — a window will warn that the trigger keywords are empty. The parameter model field isn't always visible: a new trigger sets the standard model by itself.

How to write a parameter prompt: examples for dates, IDs and names

A parameter prompt is a question to AI that it answers with a single value from the conversation. State the format and what to write if there is no value:

A short list, such as branches, can be spelled out right in the prompt: “downtown branch — 1, airport branch — 2”. The client says “hair dye”, but the directory says “Hair coloring”? Have the bot name the service as it appears in the directory and confirm it: then AI picks the right ID.

How the bot starts the search: a phrase in the conversation plan

It's most reliable to start the search with the bot's own phrase: it says it once it has all the parameters. Open the “Instruction” tab → “Plan” → the “Conversation plan” block, click “Add step” and drag the steps into place:

1. After the question about the service: “Name the service as it appears in the directory and confirm it”.
2. After the question about the day: “When you know the service and the day, write ‘Checking the schedule...’, add nothing else and wait for an answer”.
3. For a repeat search: “If the client wants another day or service, write ‘Checking the schedule...’ again”.

Keep these steps together, and put other questions before or after them. Before booking, have the bot repeat the summary: service, staff member, date and time.

How to test the dynamic database in Training AI

Open the test chat: “Training AI” → “Manual training” → “Start training” — and write as a client, naming a service and a day. When the trigger fires, a “Webhook: <trigger name>” block appears in the chat: the keyword, the “AI answer” for each parameter, the final link and the database answer.

Check the parameters against the directory. Not happy with the bot's reply after the block? “Fix” under it edits the “Dynamic webhook instruction”, not the main one. The answer stability check link repeats the request 10 times and shows whether dates and prices drift. Check the booking in the “Data write test” → “Test the data write”. Read more in “How to train an AI voice bot”.

How to review a call with a dynamic database

Did the bot name the wrong time on a real call? Open “Training AI” → “Find a call”, paste the ID from “Call details” or pick a call from the last 7 days → “Open the call”. The webhooks take their places in the conversation:

1. The init webhook block — what the bot knew before the conversation, such as the directory.
2. The dynamic database webhook block — the link, the AI sub-request parameters and the server response with its HTTP code. A wrong parameter means you should fix its prompt.
3. The smart answer block — what the bot said based on this data. Wrong slots — “Fix” under the bot's reply.
4. The write webhook block — your system's response. For example, AI wrote update instead of create, and the booking wasn't created: refine the prompt — new booking create, reschedule update, cancellation delete.

Compare against the server response saved in the call: a link opened again will already show different slots.

Common mistakes

• The trigger doesn't fire: it is turned off, or the keywords aren't in what the bot and the client actually say.
• The bot booked the wrong service: the loading webhook has no directory, or the bot didn't name the service as in the directory.
• Long silence on the line: the server responds slowly — speed up the search on your side.

Read also: AI voice bot knowledge base, How to train an AI voice bot, Turnkey AI voice bot