Connectors to your systems
Live data in the dialogue
A connector is a bridge between the assistant and your systems. It takes prices, stock and statuses at the moment of the question — from 1C, the CRM or your own API. It runs on models with function calling: the assistant picks the tool itself.
If it does not know, it checks
The assistant stops retelling yesterday’s price list. When the question touches live data, the connector asks your system during the dialogue itself.
Stock and prices
Before answering, the assistant checks 1C — the customer hears today’s number.
Order status
An order number and the assistant finds it in the CRM: where the parcel is and what comes next.
Booking and reservations
Free slots come from your calendar; the chosen time is taken immediately.
Your own systems
Any internal API: plans, limits, the customer account — within an agreed list of actions.
The path of a single call
Connectors work only with LLMs that support function calling. The model receives the tools as schemas and calls them itself — whenever the question calls for it.
The customer’s question
«How much is delivery to Kazan?» — an ordinary sentence in the chat, no commands.
The LLM picks a tool
Function calling: the model sees the schemas of allowed actions and decides to call a connector itself.
The call into your system
The connector goes to 1C, the CRM or an API — within the agreed list, with keys from the project vault.
An answer with live data
The result comes back to the model — the customer sees today’s number and the call stays in the log.
Layer by layer
A request passes six layers and comes back with live data. Only the LLM layer with function calling can start the chain: without it there is nothing to trigger a tool.
Models without function calling do not pass the third layer: they have no way to initiate a call, so connectors do not work with them. Parsewise defaults to a model with tool support.
A bridge with railings
Access to your systems is not trust on a promise. Allowed actions, limits and a log of every call: the security team has something to audit.
List of actions
A connector can only do what was allowed in advance — an agreed list of tools.
Keys in the vault
Credentials live in the project settings — they never reach the dialogue or the customer.
Call log
Every call into your system is recorded: who, when, with what and what came back.
Rate limits
A cap on requests per minute — the assistant will not overload your system.
Timeouts and cache
A slow response will not hang the dialogue, and frequent data comes from cache.
Per-project switch
A connector works only where it is switched on — convenient for starting with a pilot.
Questions about connectors
Only with LLMs that support function calling. The model receives the tools as schemas and decides when to call a connector; without that mechanism there is nothing to trigger a call.
Faster than asking again
The answer from your system arrives in a fraction of a second. And background jobs do not wait for a question — the scheduler walks through them every minute.
Action call
from the question to the answer from your system
Scheduler cycle
background jobs on a schedule
Of calls in the log
every call is visible and auditable
No duty officer
the assistant checks for itself
Question. A live answer from your system, at the moment of the dialogue.
Seconds. The scheduler step — recurring jobs run on time.
Retellings. Yesterday’s data is gone from the answers.
It answers now — and works on a schedule
The scheduler carries out instructions without being asked. Check stock before opening, chase an abandoned request, refresh statuses — the job is set in words and then runs itself.
Connect live data
Start with one connector Stock from 1C or statuses from the CRM — and the assistant’s answers stop going stale.
Sign up
Create an account in a minute — no card, free plan forever.
/02Data
Upload documents, products and profiles — the knowledge base is ready.
/03Connector
Pick the system and the credentials — the list of actions is agreed in advance.
/04Live answers
The assistant checks your system in every dialogue.
Tell us your task
Projects by type grow year over year
MVP, redesign, AI and support — cumulative
The studio profile across key axes
Speed, quality, transparency, engineering
Research, design and build overlap
Parallel streams — not a waterfall