Before releasing a bot to live clients, it is worth checking. There are several ways — from a safe dry run of the whole scenario to a point check of a single node. All of them work on a bot that already serves people too: live clients receive nothing extra.
The dry run
The scenario executes in full, but the bot’s replies are sent nowhere — they are collected into a report. You can set the client’s answer sequence in advance and see where the conversation led and what the bot replied at each step. It is a safe check: even on a live bot, not a single person receives anything.
Running from the middle
You can start from any node, setting the variables in advance. No need to walk the whole conversation from the beginning to check one edit near the end — it saves time and AI requests.
A single node run
One node executes for real: an AI node really calls the model, a web request really goes out, a script really runs — but without the whole conversation’s surroundings. The report shows the result and the chosen branch — convenient for checking a condition, a request or data extraction separately.
Pinning a step’s result
A step of a real conversation can be pinned — its result saved — and later reproduced during checks without calling the external system every time. The node under test still executes for real: pinning substitutes only what came before it, never what you are editing right now. Pins have no effect on live conversations at all — they apply only to checks.
Replaying a conversation
A recorded conversation can be run again against the new scenario version and compared with the old one: where the path and the answers changed, and where everything stayed as it was. It is the best way to make sure an edit did not break what already worked.
Tracing
Every conversation has a full trace: entering and leaving nodes, the transitions made, variable values at every step, and AI calls with their response times. The trace is visible during the conversation and after it — from the trace it is easy to understand why the bot answered the way it did.
Where problems show up
Problems are gathered in several places, but they all say the same thing: the icon and readiness verdict in the bot list, the problems list in the studio with a jump to the node, and the publish checklist behind a button. So it is always visible that the bot is not ready yet — and what exactly is worth fixing.
Checking is most convenient on the Emulator, and assembling fixes in the Flow Studio. Numbers on conversations that already happened are in the Analytics & Logs section.