How the agent works
The agent builds the pieces of Chartnaut for you from a description in plain words. Open Agent at the top of the sidebar, say what you want, and it writes the code, checks it against real market data and hands the result to you to approve.
What it can build
Ask for | What you get | Example |
|---|---|---|
An indicator | Values drawn on the chart, bar by bar | A session VWAP with ±1σ and ±2σ bands that reset at the New York session open |
Your setup, marked every time it happened in history | A London sweep of the Asia high that fails to hold | |
A study | What followed every one of those events, measured | How often New York opening-range breaks on ETH continue or fail back inside |
A study's answer on the chart as the setup forms | Attach the opening-range study to BTC | |
A page of your own, or a tab on every trade, day or report | A pre-market briefing of overnight moves and what you still hold |
It also searches your library, so "find my fair-value-gap studies" works as a request. The examples on the Agent page are a good place to start.
One request can cover several pieces. Ask for a study of London sweeps and the agent writes the definition, runs it over history, builds the study and runs it, asking you at each point that needs a decision.
Supervised and Autonomous
Next to the model picker in the chat box is a coloured capsule that sets how closely the agent checks in with you.
Supervised walks you through each step on the charts, so you can correct a mistake as it happens. The agent shows you each piece before anything uses it: a definition before it runs over history, a study before its run. New conversations start here.
Autonomous suits easier tasks the agent is likely to get right first time. It works through to your goal and shows you the finished result.
Supervised costs a click where it matters most. A definition that marks the wrong bars is caught before it runs over years of data, and before a study is built on top of it. If you change mode while the agent is working, the new mode applies from its next turn.
Approval cards
When the agent needs a decision it stops and shows a card in the chat. Nothing on the card happens until you act on it.
Card | What it asks | Your answer |
|---|---|---|
Plan | The steps it intends to take, and what done means | Approve |
Question | Something it needs to know from you | Pick an option or type an answer |
Data | The instruments, timeframe and profile a definition will run over | Approve to run it |
Profile override | Whether to change a profile that existing results depend on | Override |
Study run | The instrument, timeframe and number of events in scope | Run |
Result | The finished indicator, definition or app, with what it checked and what it did not | Accept, or Open to look first |
To ask for a change instead, type it in the chat. The card keeps your reply on record, and the agent carries on from it. Study results arrive as their own card with Open the full report, and an attached Forward Insight shows Saved to the chart. Approvals covers each card.
The builders' own agents
Each builder has an agent of its own in its chat: the indicator builder, the definition builder, the study builder and the app builder. Each one works on the single thing open in front of you, with its code and its preview beside the chat.
Use Agent for work that spans several pieces, and a builder's chat to refine one. The builders' agents explains the differences.
What it costs
Agent work uses AI credits, which you buy separately from your plan. How they are billed is covered in your account docs.
In Chartnaut Desktop you can run the agent on an AI subscription you already pay for. Open Settings → AI providers in the sidebar and connect your Claude subscription, your ChatGPT subscription through Codex, or both. Chartnaut's AI then runs on that subscription instead of credits, and the page shows how much of its 5-hour and weekly allowance you have used. Use your own AI subscription walks through it.
Codex