Understanding Falcon AI
How a Falcon AI conversation works, and what each turn draws on
A conversation is a thread, a turn is one exchange
A conversation is one thread with Falcon inside a workspace. A turn is one thing you ask plus everything Falcon does to answer it: reading your message, deciding what to use, calling tools if it needs to, and writing a response. A conversation is a sequence of turns; everything below describes what happens inside one of them.
What a turn draws on
Every turn draws on four things:
- Your message, and any file attached to it. The question or instruction you typed, plus anything you uploaded alongside it
- The page you asked from. Falcon knows what part of the platform you were looking at when you asked, the same way it knows which evaluation you mean when you ask about one without naming it
- The skill that’s active. A skill is a packaged, repeatable workflow, built-in or custom, that Falcon follows with the right tools already loaded. If one is running this turn, it’s carried along with the turn, the same as your message or the page
- The tool set Falcon loads for that turn. Tools are what let Falcon look things up and make changes on the platform, rather than just describe them; the pool includes any external tools you’ve connected through MCP Connectors alongside the built-in platform ones, and Falcon loads only a working subset of it for each turn
Falcon picks that subset by reading your request and settling on a working area, roughly forty tools brought forward for the turn out of the hundreds it could load. How specific your ask is decides how wide that area gets:
- A vague ask keeps the working area broad and the tool set wide
- A specific ask narrows both
You can also point Falcon at a working area directly, either by opening the context selector in the chat input and picking one, or by naming the area in your message.
flowchart TD accTitle: The four inputs a Falcon AI turn draws on accDescr: A conversation contains turns. Each turn draws on your message and any attached file, the page you asked from, the active skill, and a tool set. The tool set is a working subset of the full pool of tools Falcon could load. CONV["Conversation · one thread in a workspace"] --> TURN["Turn"] TURN --> MSG["Message + attached file"] TURN --> PAGE["Page you asked from"] TURN --> SKILL["Active skill"] TURN --> TOOLS["Tool set for this turn"] POOL["Full pool of loadable tools"] -->|"~40 brought forward"| TOOLS
Watching a turn run
While a turn runs, the answer streams in as it’s written. Each tool Falcon calls shows up as its own card: it starts with Running…, and once the call finishes you can open it to see the Parameters it was given, the Result, and the Full output behind that result. If the turn creates or changes something on the platform, a completion card appears at the end with a link straight to it.
Note
As a conversation grows very long, Falcon automatically condenses the earlier part into a summary so the thread stays usable, while the most recent turns are kept exactly as written. Falcon still draws on what was discussed early on, but only as that summary rather than the original wording, so if an exact detail from far back in the thread matters, restate it rather than assume Falcon recalls it precisely.
Cost and pace
Each turn costs one AI credit. You’re also capped at ten messages a minute; go past it and Falcon shows an error in the conversation asking you to wait before sending more. If your organization’s AI credit balance runs out, the turn is refused with an error in place of an answer. Credit balance is tracked on Billing & Pricing.
Why it matters
The page you ask from, the skill that’s active, and how specific your ask is all shape the tool set Falcon brings forward for a turn, and that shapes the answer you get back. That matters most in a long back-and-forth, where the message limit and credit cost add up turn by turn.
Keep exploring
Questions & Discussion