AI Agent: a bot from a description
The built-in AI Agent builds, edits, explains and fixes scenarios from a plain-language request. It opens as a separate floating window and works alongside the editor. On Free it builds from the free blocks, on PRO from all of them.
Build the graph in conversation
The agent knows the active scenario. It explains features, creates and edits nodes on demand, and helps trace why a run misbehaves.
- Scenario-aware answers
- Create · edit nodes
- Error diagnosis
What it can do
Builds a bot from a description
“Farm resources, and when the bag is full, go sell” becomes a finished graph with conditions, retries and pauses.
Asks first
When a choice really changes how the bot behaves — how to recognise the target, what to do on failure — the agent asks one simple question. It settles the small stuff itself.
Builds to a plan
Work goes in stages, and each stage is checked before the next begins. A failure rolls back only the unfinished stage — finished work is kept.
Fixes instead of looping
The agent understands what exactly went wrong and never repeats the same failed attempt. If the task cannot be solved, it leaves a working draft and explains what to finish by hand.
Only what you actually have
Windows, image templates and models come from what is really set up. If something is missing, it names the setup step instead of making things up.
Sees images
Paste a screenshot from the clipboard or upload a file — the agent understands what is on screen. Needs a model that accepts images.
Knows the whole program
MultiBox, Sandbox, Image Lab, text recognition, game memory. Reads the run log and explains why the bot stumbled or a copy failed to start.
Ready-made recipes
Common patterns, such as a farm loop, are assembled in one step and then tuned to your game.
Changes stay under your control
- Every graph edit is validated: the scenario must stay runnable and links must stay compatible.
- Changes to MultiBox rules are only proposed: you see the exact change and approve or reject it. A proposal expires after a while.
- Agent messages can be copied one by one or all at once.
Provider, model and cost
- Anthropic, OpenAI, DeepSeek and any service with an OpenAI-compatible API are supported — your own server or a local model, for example.
- Paste your API key and the agent loads the list of models your account can use.
- Input and output tokens per request can be capped.
- A session counter in the agent window shows how many tokens were used and what that cost in dollars. Prices for known models are built in and can be adjusted if the provider changes them.
Click the AI button in the titlebar to open the floating chat window. You can drag it outside AIWeavio entirely, resize it by the edges and keep it visible while you edit the graph. The main window keeps running at full speed — one doesn't slow the other down.