Paxy AI

Paxy - Your AI assistant who can answer questions about your data.

Overview

Paxy is the AI reporting assistant inside PAX. It helps authorized users ask plain-English questions about ERP and CRM data, then returns a concise answer with supporting findings and export options.

Paxy is meant for questions like:

• "What were our total sales to CUST-001 in 2025?"
• "When is the last time we received PART-100(REV-A)?"
• "What have our purchases to VEND-001 been so far in 2026?"
• "What is currently on hand for this part?"
• "Which customers ordered this item last year?"
• "What is the average cost to make this part?"

The goal is not to replace every formal report in PAX. The goal is to make data questions faster to answer, especially when the user knows what they are looking for but does not want to build a report manually.

Who Paxy Is For

Paxy is currently for admin and executive users.

It is designed for people who need cross-functional visibility across sales, inventory, purchasing, manufacturing, CRM, and accounting data. The feature is intentionally restricted because it can access broad business data.

Users without the required role do not see the Paxy launcher and cannot use Paxy.

What Paxy Does

Paxy takes a question, finds the relevant area of the system, creates a read-only report, and summarizes the result.

It can help with:

• Sales and revenue questions
• Customer order history
• Invoice-based sales totals
• Purchasing and vendor activity
• Receipts and purchase order history
• Inventory, lots, availability, usage, and part details
• Manufacturing and work order questions
• Quotes, shipments, returns, and CRM activity
• Accounting-related summaries when the question requires them

Paxy returns:

• A short answer
• Key findings
• Notes when needed
• A CSV export with the supporting data
• A PDF findings report

The chat answer does not show internal report details or a large data table. The detailed rows are available through the CSV, and the PDF gives a cleaner summary report.

How To Use Paxy

1. Open PAX as an admin or executive user.
2. Click on Paxy the Penguin on the left side of the window.
3. Type a question in the chat window.
4. Press Enter or click Send.
5. Wait for Paxy to respond.
6. Use CSV when you want the detailed export.
7. Use PDF when you want a short findings report.

You can close the chat while Paxy is working. When the answer is ready, the Paxy launcher shows a red notification dot. Reopen Paxy to see the result.

The chat remains available while the current page session stays open. If the page reloads or the app is reopened, the current chat history is not intended to persist.

Tips For Better Questions

Paxy works best with precise business identifiers.

Use exact values when you have them:

• Customer codes, such as CUST-001 or CUST-002
• Vendor codes or precise vendor names, such as VEND-001 or Sample Vendor
• Part numbers, such as PART-100(REV-A) or MFG-PART-001
• Order, invoice, quote, PO, or work order numbers
• Clear date ranges, such as "in 2025", "so far in 2026", or "last 90 days"

Good questions:

• "What were total invoice sales to CUST-001 in 2025?"
• "When did we last receive PART-100(REV-A)?"
• "Show purchases from VEND-001 so far in 2026."
• "What customers bought MFG-PART-001 in 2026?"

Less ideal questions:

• "How are we doing?"
• "Show me that vendor thing."
• "What happened with that part?"

Paxy can sometimes still answer broad questions, but precise identifiers reduce confusion and improve the quality of the result.

Current AI Models

For now, Paxy uses:

• gpt-5.4 for understanding the question and preparing the report.
• gpt-5.4-mini for summarizing the findings clearly and preparing the PDF report content.

This setup balances quality and speed. The heavier reasoning happens before the data is retrieved, while the final response is kept concise and practical.

What Happens Behind The Scenes

When a user asks Paxy a question, the process is:

1. PAX checks that the user is logged in and has the admin or executive role.
2. PAX identifies which business areas are relevant, such as sales, inventory, purchasing, manufacturing, or accounting.
3. PAX prepares a read-only report for the matching data.
4. PAX validates that the report can only read data and cannot change anything.
5. PAX retrieves the matching rows and applies result limits.
6. PAX formats timestamps for the user's timezone when possible.
7. Paxy returns a short answer, findings, notes, and export-ready data.
8. The user can download CSV or PDF from that same result.

If no matching rows are found, Paxy simply returns a no-results message.

Security Measures

Paxy has several layers of protection.

Role-Based Access

Only admin and executive users can access Paxy. The launcher is hidden for other users, and PAX also blocks unauthorized access behind the scenes.

Tenant Isolation

Paxy follows the same tenant protections as the rest of PAX. Requests are checked against the logged-in user's tenant, so one tenant cannot access another tenant's data.

Read-Only Reporting

Paxy is only allowed to read data for reporting. It cannot create, update, delete, post, void, or otherwise change records in PAX.

Before PAX retrieves results, it checks that the report request is read-only and safe to run.

Result Limits

PAX caps results before returning them.

Current limits include:

• Up to 500 rows in the detailed export result
• Up to 50 columns in the detailed export result
• Up to 50 rows in the PDF table
• Additional limits before result rows are summarized by AI

This keeps Paxy responsive and helps prevent overly large accidental reports.

Sensitive Field Filtering

PAX does not include known sensitive fields in the data context sent to the AI model.

Examples of fields excluded from AI context include:

• Passwords
• User session data
• API keys and tokens
• Employee PINs
• Raw receipt attachment data
• FedEx label data
• Tenant logo binary data
• Email HTML body/template content
• Email template attachments

PAX also filters private fields again before result rows are sent to the summary model.

Protected AI Connection

PAX handles the AI connection securely behind the scenes. Users do not need to manage keys or connect directly to the AI provider.

If Paxy is not configured, the user sees a setup message instead of a broken chat experience.

CSV And PDF Do Not Re-Ask The Model

The CSV and PDF buttons use the result Paxy already generated. They do not make another AI request and do not reinterpret the question.

Data Sent To The AI Model

Paxy sends only what is needed for the reporting workflow:

• The user's question
• A filtered description of the relevant data areas
• The allowed result rows needed for the final answer
• Result metadata, such as row counts and whether rows were capped
• The user's browser timezone, when available, so timestamps can be presented clearly

Paxy does not send the full database by default. It uses curated data areas so the model sees the most relevant information for the question.

Timezone Handling

When Paxy receives timestamps, PAX formats them using the user's browser timezone when a valid timezone is available. This means answers like "Jan 15, 2026 at 8:30 AM" are intended to match the user's local view of time, not a forced UTC display.

CSV And PDF Outputs

CSV

The CSV is best when the user wants the supporting data. It includes the capped result rows and columns returned for that answer.

PDF

The PDF is best when the user wants a cleaner report to review or share. It includes:

• Report title
• Original question
• Answer
• Findings
• Notes when needed
• A small supporting table when useful

The PDF is intentionally capped so it remains readable and does not become a massive raw data dump.

What Paxy Does Not Do

Paxy does not:

• Change database records
• Create orders, invoices, payments, or inventory adjustments
• Replace formal accounting review
• Guarantee that a broad or vague question was interpreted exactly as intended
• Keep a saved history of every report run
• Store long-term chat history

Paxy is a reporting assistant. It should be used as a fast way to inspect and summarize data, with normal business review for important decisions.

Best Practices

For the best results:

• Use exact part numbers, customer codes, vendor codes, and order numbers.
• Include date ranges when time matters.
• Say whether you want invoice sales, order totals, open orders, receipts, or purchases.
• Ask one focused question at a time.
• Use CSV when you want to audit the underlying rows.
• Use PDF when you want a concise findings report.

If a result looks surprising, use the CSV export to review the source rows behind the answer.