Separate the model from the tools it can use
A model's name does not tell you what a trading system can do. Inspect the tools, account permissions, and workflow around it. A plain conversation can help organize research; a connected application may also retrieve current data or call an external API.
For example, OpenAI documents custom GPT actions that connect to external APIs. Calling every custom GPT “just a persona” is inaccurate. What matters is the configured action, its authorization, and the data actually supplied to the model.
Three implementation paths
A research assistant and manual review
Ask the model to organize a brief, list missing information, or explain a rule. Verify the cited source and its timestamp, then decide what to do yourself. Treat any unsourced price or account balance as unverified. This path is useful for refining a specification before building automation.
A model API inside your own program
Your code handles market data, account access, validation, scheduling, order state, retries, and logs. A model response is one input to that program. Require an explicit schema and reject invalid or incomplete responses before they reach an action. Estimate engineering effort from those responsibilities rather than from the number of lines in a model-call example.
An inspectable workflow in NickAI
Describe the task to Nick and inspect the resulting nodes. You can combine price or stock data, Function and Conditional nodes, a model analysis step, notifications, and an account connection. The workflow assistant guide explains the builder; the LLM node reference covers model configuration.
A first workflow: a stock-watchlist briefing
Start with a research task: “Create an inactive daily briefing for my stock watchlist. Show the data source and timestamp, summarize the inputs, and send a report. Ask me for the watchlist, timezone, and destination. Include no order node.” Review the request in Nick's composer, then send it when it describes your task.
- Confirm the assets, data provider, and lookback in the data node.
- Inspect the model prompt. Ask it to distinguish source observations from interpretation and to report missing data.
- Run a manual test and compare the briefing with the source output.
- Check the destination and notification result, then set a supported schedule.
For a larger first-party example, inspect the Space Stocks Signal Monitor. Its public template shows the data, research, scoring, and reporting stages. Open the template to review the actual configuration. This article describes the setup process; it does not claim a measured result from that template.
What about Claude and MCP?
Using Claude as a model inside a NickAI workflow is different from connecting Claude Code to an external server. Claude Code supports MCP servers, whose available tools determine what the client can do. Our MCP guide shows how to inspect that boundary.
An HTTP GET check of the previously documented NickAI public MCP route returned 404 on September 7, 2026. A supported external connection was not verified. This guide therefore uses the in-app builder for its reproducible NickAI path. Confirm a current supported endpoint before following an older external-client setup.
Expand to a paper workflow after the briefing works
When inputs and reporting are clear, use the Alpaca workflow walkthrough or a paper strategy example. Inspect every account destination and test each decision branch. Multiple models can expose disagreement, but their agreement is not evidence that a decision is correct.
Can ChatGPT trade for you?
Not on its own. A chat model reads what you give it and writes text back; it holds no exchange account and sends no orders. Trading needs three more parts: a market data feed, rules that turn the model's output into a decision, and an execution connection that submits the order.
What a chat model does well: summarize news, explain a chart you describe, draft a strategy rule, and check a plan for missing conditions. What it cannot do alone: watch a market between messages, confirm that a price it quotes is current, run on a schedule, or place and track an order. Any tool you connect adds a capability, so check the tool list and permissions rather than the model name.
ChatGPT trading bot vs an agentic workflow
The difference is who supplies the missing parts. In a chat, you do. In a custom GPT or your own script, your code does. In a NickAI workflow, nodes do, and you can inspect each one.
| Capability | ChatGPT chat | Custom GPT script | NickAI workflow |
|---|---|---|---|
| Market data | What you paste, or what a browsing tool finds | Whatever API your action or code calls | Built-in data nodes for prices, stocks, news, Kalshi and Polymarket |
| Execution | None: you place the order yourself | Your code calls the exchange API with your keys | Order nodes for the venues on the integrations list |
| Scheduling | None: it runs when you type | Your own cron job or server | Built-in schedules and webhook triggers |
| Guardrails | Your own judgment | The checks you write and maintain | Conditional and Function nodes, inactive drafts, a paper exchange |
| Code required | No | Yes | No: described in plain language |
For how model-driven workflows differ from fixed-rule bots, see agentic trading vs algorithmic trading.
Building a ChatGPT trading bot step by step
- Write the rule in one paragraph. Name the asset, the input, the condition, the action, and the largest order size allowed.
- Pick the data source. Choose the feed for each input and record its timestamp. A price without a time is not usable.
- Give the model one job. Ask it to classify or summarize and to answer in a fixed format. Check that output with a plain rule before any action. The LLM evaluation guide covers how to compare models on that job.
- Connect execution in paper mode. Use a paper account first, such as an Alpaca paper account or the NickAI paper exchange, and select it in every order step.
- Schedule, then review the log. Run it on a timer, read each run's inputs and decisions, and go live only after both the action and the no-action branches behaved as written.
In NickAI, you describe steps 1 to 5 to Nick and it builds the workflow as an inactive draft for you to review. See the automated trading agent use case for a worked example.
What about forex or binary options?
NickAI connects to the venues listed on getnick.ai/integrations: crypto exchanges, Alpaca for US stocks, and the Kalshi and Polymarket prediction markets. A forex pair is only reachable if one of those venues lists it, so check the list first. NickAI does not connect to binary options brokers, so a binary options bot is outside what it builds.