Think with AI

AI steps analyze, organize, or create content from the information in your workflow. Choose the kind of work, its instructions, and the data it receives. Free, Pro Lite, and Pro can use AI steps within their plan's limits and available model choices.

Before you begin

Open a workflow and add any data sources the analysis needs.
Keep sufficient GNG credits available, including for tests.
Create a saved subagent first if you intend to use Run my subagent.
Confirm that your account also has War Room access before using War Room council. Workflow access alone does not include council access.

Quickstart

Add AI Analyst under Think with AI.
Select an available model or GNG Analyst for automatic model selection.
Write a Prompt that names the question, required evidence, and desired answer format.
Connect the data steps the analysis should use.
Review Data and Tools, then select Test.
Inspect the answer and GNG credits in the run before saving the workflow.

What it shows

Step	Purpose
AI Analyst	Analyze supplied information and use selected research tools.
Structured extract	Return named fields for later comparisons or outputs.
AI judge	Answer a yes/no question about the supplied data.
Formatter	Turn earlier results into the requested written format.
AI image	Create an image from a written or selected prompt.
AI email writer	Compose an email subject and designed body for Email me.
Run my subagent	Apply a saved specialist's model and instructions to a task.
War Room council	Convene a panel and return its synthesis and replay.

Inputs

Model and reasoning

Select from the models offered for the step and your plan. Where a Reasoning control is shown, it offers Low, Medium, High, and Max. Deeper reasoning generally uses more GNG credits. Model names and availability can change; use the current picker.

AI Analyst instructions

In Prompt, choose Write inline or a saved prompt. To reuse an inline prompt, enter a name beside Save to library and select that action. Selecting a saved prompt displays its preview. A missing saved prompt must be replaced or rewritten inline.

Analyst persona (optional) adds standing instructions about the analyst's approach. Keep the question and requested deliverable in the prompt.

Data and tools

Include memories adds relevant AI Console memories.
Choose specific fields lets you select results from earlier connected steps, including information behind a condition. Use Add data field for each selection.
Continue from a chat (optional) supplies context from one of your saved chats. Incognito chats are unavailable. Choose None to remove it.
Tools selects the research capabilities the analyst can use. Choose those required by the task.

By default, the AI receives results from directly connected steps. A condition directly before the AI supplies its yes/no result. Select earlier news, quotes, or holdings explicitly if the analysis needs them.

Advanced analyst controls

Max output tokens (optional) limits response length, and Max tool rounds (optional) limits repeated tool work. Follow the displayed model guidance; a very small response allowance can prevent a useful answer.

Temperature (optional) appears for compatible models and adjusts variation. Web search appears where supported and permits current web research with additional GNG credit usage.

Structured output returns named fields rather than unrestricted prose. It requires a specific model and disables tools, web search, and temperature. In the field builder, select Add field, enter a unique name, choose Text, Number, Yes-no, or a list type, and set Required as needed. Add descriptions to explain what each field should contain.

Extract, judge, and formatter

Structured extract uses a Prompt and the same field builder. AI judge takes a Question and returns a true/false verdict that a later Condition can compare. Formatter takes Instructions describing the final written format. These steps also offer model, reasoning, and specific-data choices.

Images and email writing

AI image needs a Prompt or an earlier Prompt source. A connection alone does not supply the image description. Select Size as square, portrait, or landscape. Connect the resulting image to the output that should use it.

AI email writer takes What the email should say, a model choice, and selected data. Allow images lets it include connected chart or image results. The writer creates the message; connect it to Email me and select AI designed to send it.

Saved subagents

In Run my subagent, select Subagent and optionally enter Task. The saved model, reasoning, instructions, and memory choices remain in use. Include memories adds your general AI Console memories. Review unavailable or disabled-agent warnings before running.

Workflow tools can differ from the tools that subagent uses in chat. Check the displayed count, such as X of Y tools run in workflows, and account for your plan's permissions. If a needed tool is unavailable, supply the required information through connected data steps and select the relevant fields before testing.

War Room council

Your account must also have War Room access. Enter a Topic, or enable From upstream field and select Topic source. Build the council with Add saved seat, Add custom seat, or Load saved lineup.

To generate a council, write the Topic inline and leave From upstream field off, then select Design council with AI. If seats already exist, confirm their replacement before generation starts. Generation uses GNG credits and immediately replaces the draft's seats, mode, and round style. Review those changes before Test and Save.

A custom seat has Name, Instructions, Model, reasoning, and optional selected tools. Council settings include Mode, Round style, Rounds, Optimal model mode, Orchestrator, Enable all tools, and Reasoning. A loaded lineup can be reduced to the choices permitted for the workflow; read any notices.

Set Spend cap in GNG credits. Feed connected data into the council briefing includes the supplied research, while Choose specific fields narrows what the panel sees. Review the synthesis and replay in the run output.

Outputs and how to read them

Open the AI step in a test run and inspect Output. Check whether it answered the requested question and received the required sources. A structured result should contain the named fields later steps expect; an AI judge's verdict is a decision, not the underlying evidence.

How it is calculated

GNG credit usage depends on the selected model, reasoning, tools, and work performed. Run history shows the overall amount and each step's cost. AI-assisted workflow creation, style generation, and diagnosis also use credits.

Data and timing

Instructions and selected data are applied when a step runs. A saved chat contributes summarized context rather than a guarantee that every earlier message is included. Use explicit data selections for facts the task must receive.

Use cases

Apply a saved specialist to each recurring portfolio review.
Convert news into named fields before evaluating a condition.
Ask a council to compare opposing interpretations of a market move.

Limitations and disclosures

AI can omit information or reach an incorrect conclusion. More reasoning does not guarantee accuracy. A successful step can still produce a weak answer if it receives only a condition result instead of the relevant research.

GNG Research provides equity research and educational tools, not investment advice. Nothing on the platform is a recommendation to buy or sell any security. Do your own research and consider your circumstances before making any investment decision.