> For the complete documentation index, see [llms.txt](https://docs.coherent.global/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://docs.coherent.global/assistant/agent/ai-guidelines.md).
# AI Guidelines
{% hint style="info" %}
Agent uses AI features that may still be improving. Availability, model options, and usage limits can vary by organization.
Contact [Support](/support/faq.md) if you need help with access, usage limits, or your organization's AI requirements.
{% endhint %}
Agent helps you review workbook context with AI while keeping you in control.
Agent helps you work with Excel through natural-language requests. You can ask it to explain formulas, summarize sheets, create analysis, compare files, review workbook risks, and make approved workbook changes.
Because Agent uses AI, treat its responses as a helpful assistant's work rather than a final control. Review important outputs, confirm business assumptions, and keep your organization's data rules in mind.
## How Agent uses AI
Agent combines workbook context from your device with the instructions you provide in chat. It decides what information is relevant, prepares a focused request, and sends that request to the language model available to your organization.
Agent does not upload an entire workbook just because you open Agent. It uses local workbook information, selected ranges, attached files, conversation history, and saved instructions only when they are relevant to your request.
## Good uses for Agent
Agent works best when the task has a clear goal and a clear source of context. For example, you can ask Agent to:
* Explain the purpose of a workbook, sheet, range, or formula.
* Identify key inputs, outputs, assumptions, and calculation flow.
* Review formulas for inconsistencies, hardcoded values, external links, or errors.
* Summarize table-style data by region, month, customer, product, or status.
* Create or adjust workbook content after you approve the change.
* Compare workbook versions and summarize material differences.
* Use an attached file or knowledge base article as background for a task.
For broad work, start with a review request before asking Agent to make changes. A short summary often reveals which follow-up questions are worth asking.
## Write requests that give Agent enough context
A strong request tells Agent what to do, where to look, and how to answer.
Use this pattern:
> Do \[task] using \[sheet, range, workbook, attachment, or article]. Return \[format, level of detail, or decision criteria].
Examples:
* "Review `Model!A1:K120` for formula consistency. Return the five highest-priority issues with cell references."
* "Summarize the attached workbook for an approver. Focus on assumptions, outputs, and unusual changes."
* "Use the selected table to create a revenue summary by month and region. Show the result before editing the workbook."
Avoid requests such as "fix this" or "analyze everything." If a request is too broad, Agent may ask a follow-up question or produce a shallow answer.
## Review AI responses before relying on them
AI responses can contain mistakes, miss context, or overstate confidence. Review Agent's work before using it for business decisions, customer communication, regulatory reporting, or production workbook changes.
For higher-risk tasks:
* Ask Agent to show cell references, assumptions, and reasoning in a concise form.
* Ask for a preview before workbook edits.
* Use read-only mode for review and Ask for edit mode for changes.
* Check formulas and outputs after Agent changes a workbook.
* Save a copy of important workbooks before major edits.
## Data handling and privacy
Agent is designed to use local workbook information first. Conversations, saved outputs, Skills, and temporary files are stored in browser storage on your device. When Agent needs an AI response, it sends only the information needed for that request.
The original workbook file is not shared with the AI provider. Agent extracts and sends only the information necessary to perform the requested analysis.
Coherent uses OpenAI through API business terms, where customer prompts and outputs are not used to train OpenAI models by default. OpenAI's standard API retention is limited, commonly up to 30 days, and is primarily for abuse monitoring and service operations, after which the data is deleted unless a legal requirement applies.
The exact model options and AI provider configuration can vary by organization. Use the settings available to your organization, and contact Support if you need details about your approved configuration.
Do not include sensitive information in prompts unless your organization allows that use. If your team has stricter data handling requirements, confirm the approved workflow with your administrator or contact Support.
## Limitations
Agent is useful for workbook assistance, but it has practical limits:
* Very large workbooks or attachments may take longer to process or may need narrower requests.
* Complex tasks may require follow-up prompts, smaller ranges, or separate steps.
* AI can misread ambiguous labels, hidden assumptions, or business-specific context.
* Agent may ask for approval before workbook changes, especially when existing content could be overwritten.
* Some features, model options, and usage limits may differ by organization.
## Related pages
* [Getting started with Agent](/assistant/agent/getting-started-with-agent.md)
* [Chatting with Agent](/assistant/agent/chatting-with-agent.md)
* [Agent modes and approvals](/assistant/agent/agent-modes-and-approvals.md)
* [Settings and storage](/assistant/agent/settings-and-storage.md)
---
# Agent Instructions
This documentation is published with GitBook. GitBook is the documentation platform designed so that both humans and AI agents can read, navigate, and reason over technical content effectively. Learn more at gitbook.com.
## Querying This Documentation
If you need additional information that is not directly available in this page, you can query the documentation dynamically by asking a question.
Perform an HTTP GET request on the current page URL with the `ask` query parameter:
```
GET https://docs.coherent.global/assistant/agent/ai-guidelines.md?ask=
```
The question should be specific, self-contained, and written in natural language.
The response will contain a direct answer to the question and relevant excerpts and sources from the documentation.
Use this mechanism when the answer is not explicitly present in the current page, you need clarification or additional context, or you want to retrieve related documentation sections.