Documentation / Knowledge
Live data sources (menu, price list, stock)
Connect a URL you own; the agent reads the current snapshot at the start of every call.
Last updated:
A data source is a URL you own (a menu, a price list, stock, open slots) that the platform fetches on a schedule, turns into plain text and places in front of the agent when a call starts. There is no lookup during the call: the agent answers "which is the cheapest crêpe?" from the whole list, from the first word.
When to use it#
- The knowledge base is for documents that change rarely and are searched. A catalogue that changes daily and has to be read as a whole (prices, what comes with what) fits a data source better.
- A tool fetches one answer while the caller waits. Use a tool for per-customer lookups (an order, a booking) and a data source for the list everyone asks about.
Connect a source#
- Open the agent and go to its Data sources tab.
- Click Add data source and answer Where does the data come from? by picking your provider. iotSoft POS is first in the list; if your provider is not there, use Catalogue JSON API or Text or CSV page.
- Enter the URL and, if your provider needs one, the Access token. The token is stored as a secret on the project and is never shown again.
- Set Refresh every (minutes) (minimum 5, default 60) and, if needed, Max characters (default 15000).
- Click Connect and fetch. The first fetch runs immediately, so you see at once whether the address and the token work.
- Attach the source to the agent, then Save and Deploy.
Preview what the agent sees shows the exact text the agent gets.
How it stays fresh#
Every source is fetched in the background on its own schedule, by exactly one server at a time; a call never waits for a fetch. A fetch that finds nothing new leaves the stored data untouched and only notes when the source was last checked, so the list shows both checked and changed times. A failed fetch keeps the last good snapshot in use and marks the source error with the reason; nothing changes mid-call. Refresh now fetches outside the schedule.
Describing your own JSON#
With Catalogue JSON API, the Mapping (JSON) tells the platform where things live in your feed, in dot notation: the list of categories (list_path), the category name, the items, each item's name and price, its contents and its option groups. Free groups become the item's contents or a choice the customer makes; priced groups become extras. Adjust the mapping until the preview reads the way a customer would hear it.
Good to know#
- Up to 8 sources per agent.
- A feed larger than the cap is cut at a category boundary and the text says the list continues, so the agent never claims an item does not exist.
- Addresses on private networks are refused.
- Deleting a source also deletes its stored token.