Function calling
A model that acts
Function calling is how you give an LLM hands. The platform’s tools are declared to the model as schemas, and it calls them itself — whenever the conversation asks for it.
What it calls
Three main families of tools. Actions in the chat, live data from your systems — and any logic of your own, declared by a schema.
From a sentence to a call
The customer never sees the machinery — only the result. Inside runs a strict cycle: pick the tool, validate the arguments, call, answer.
Choosing the tool
The model works out that this needs a warehouse connector call, not a retelling of the base.
A call with a schema
The arguments are assembled against the declared schema and validated before the call.
The result in the answer
The data comes back into the dialogue and the model answers with it.
A standard protocol, not a home-made one
DECLARATION
Tools are described by schemas
Every function has a name, arguments and types. The model sees only the declared tool surface.
VALIDATION
No call leaves with broken data
Arguments are validated against the schema before the call — an error goes back to the model, not to your system.
TRAIL
Every call leaves a record
Tool, arguments, result and time — the log is available to your team in the admin.
Schema
functions are declared as schemas — the model calls them itself
Buttons
the customer writes a sentence — the model starts the action
Validation
arguments are validated before the call
Fallback
without tool calling, prompt mode takes over
Log
every call is recorded with its result
Calls on the clock
A tool must not slow the conversation down. The choice is instant, the call is controlled, and a slow response never hangs the dialogue.
Choosing the tool
the model makes the decision on the fly
Argument validation
no call leaves with broken data
Modes
tool calling and the prompt fallback
In every dialogue
the tools are always available
What the model can call
The set grows with the platform. Everything declared is available to the model in every dialogue.
Skills
Forms, bookings, calculators — actions in the chat window.
Connectors
Live data from 1C, the CRM, the calendar and your API.
Base search
RAG context: knowledge fragments straight into the prompt.
Agent
Handing the dialogue to a human when that is the right call.
Scheduler
Deferred and recurring jobs on a schedule.
Your functions
Describe the schema and the model starts calling your own logic.
Give the model tools
An assistant that acts sells better than one that advises. Switch on a skill or a connector and watch the calls at work.
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