Error Sense AI

Learn how to customize AI-powered error explanations with domain-specific instructions at global, integration, and action levels

Error Sense AI automatically analyzes and explains why your actions fail. When an action encounters an error, Error Sense provides a plain-language explanation of what went wrong and how to fix it.

How It Works

When an action fails, Error Sense AI:

  1. Analyzes the error response from the integration
  1. Generates a short summary for quick understanding
  1. Provides a detailed explanation with root cause analysis and suggested fixes

Customizing with AI Instructions

Out of the box, Error Sense uses general knowledge about integrations. To make explanations more relevant to your organization, you can add AI Instructions - domain-specific context that guides the AI's analysis.

Instruction Hierarchy

Instructions are organized in three levels:

1. Global Workspace

  • Applies to all actions across your workspace
  • Use for: naming conventions, data patterns, general context
  • Example: "Our customer IDs always start with 'CUST-' prefix"

2. Integration-Level

  • Applies to all actions within a specific integration (e.g., all HubSpot actions)
  • Use for: API quirks, rate limits, field mappings
  • Example: "In HubSpot, we use custom property 'revos_id' as the primary identifier"

3. Action-Level

  • Applies to a specific action (e.g., "HubSpot - Upsert Contact")
  • Use for: field requirements, validation rules, action-specific troubleshooting
💡

The AI receives instructions from all applicable levels. More specific instructions complement broader ones.

When to Add Instructions

Add instructions when:

  • AI explanations miss domain-specific context (custom fields, naming conventions)
  • Same errors repeat and need consistent troubleshooting guidance
  • Your integration has known quirks or rate limits the AI doesn't know about
  • Error resolutions require internal knowledge (contacts, procedures, workarounds)

Signs you should edit existing instructions:

  • You clicked Dislike on an explanation
  • AI suggestions don't match your actual workflow
  • Troubleshooting steps are outdated or incomplete
  • New error patterns emerged that need documentation

You don't need instructions when:

  • AI explanations are already accurate and helpful
  • Errors are self-explanatory from the raw error message
  • The issue is a one-time occurrence unlikely to repeat

Managing Instructions

Requirements:

  • Admin role required to create, edit, or delete
  • Growth or Enterprise plan

Navigate to Agents > Error Sense to view and manage all instructions.

Error Sense page
Error Sense page

Add Instruction

Level
Steps
Global
Click "Add global workspace instruction" at the bottom
Integration
Click "Add Instruction for [Integration]" → Select "Common instruction"
Action
Click "Add Instruction for [Integration]" → Select the specific action

Edit Instruction

  1. Find the instruction card
  1. Click the Edit (pencil) icon
  1. Modify content in the markdown editor
  1. Click Save

Delete Instruction

  1. Find the instruction card
  1. Click the Delete (trash) icon
  1. Confirm in the dialog
📝

Changes take effect immediately for future error explanations.

Writing Effective Instructions

Best Practices

Be specific and actionable:

Good: "Contact email must match format: name@ourcompany.com.
      Check the source system for malformed emails."

Bad: "Make sure emails are correct."

Include common error patterns:

Common issues:
- 'DUPLICATE_VALUE' - contact exists with different email
- 'INVALID_PROPERTY' - missing custom field in HubSpot

Provide troubleshooting steps:

If sync fails with 'RATE_LIMIT', wait 10 minutes and retry.
Our HubSpot limit: 100 API calls per 10 seconds.

Use markdown formatting:

## Required Fields
- `email` - Must be unique
- `company_id` - Must exist in CRM

## Common Fixes
1. Check source data quality
2. Verify API credentials

Recommended Content by Level

Level
Content
Global
Naming conventions, data formats, troubleshooting contacts
Integration
API quirks, rate limits, authentication notes, field mappings
Action
Field requirements, validation rules, specific error resolutions

Viewing Error Explanations

When an action fails, open the Error Explanation Dialog by clicking Explain with AI.

Where to Find

Option 1: Table view

  1. Navigate to a table with actions
  1. Find a failed action run
  1. Click Explain with AI
Explain with AI in table
Explain with AI in table

Option 2: Execution drawer

  1. Click on a row with error message to open the executions drawer
  1. Click Explain with AI
Explain with AI in drawer
Explain with AI in drawer

Option 3: Dashboard

  1. Open Dashboard
  1. Find a failed action run
  1. Click Explain with AI
Explain with AI on dashboard
Explain with AI on dashboard

AI Error Explanation Dialog

AI-generated analysis including:

  • Error Summary
  • Cause Analysis
  • Resolution
AI Error Explanation tab
AI Error Explanation tab

Raw technical data for debugging:

  • Input - Data sent to the action
  • Error - Raw error response from the integration
Execution Context tab
Execution Context tab

Feedback Buttons

Feedback buttons
Feedback buttons
Button
Description
Like (thumbs up)
Rate positively to help improve AI quality
Dislike (thumbs down)
Rate negatively and get options to improve instructions
Edit (pencil)
Quick access to create/update instructions
Reload (refresh)
Re-generate explanation after updating instructions

Improving with Dislike Feedback

Clicking Dislike opens a panel to improve the AI:

Dislike feedback panel
Dislike feedback panel

Choose which instruction to create or update:

  • Global instruction - Workspace-wide guidance
  • Integration instruction - All actions in this integration
  • Action instruction - This specific action

After saving, the explanation automatically reloads with improved results.

Edit Menu

Click Edit (pencil) to quickly access instruction management without leaving the dialog:

Edit instructions menu
Edit instructions menu

Instruction Editor

When creating or editing, a drawer opens with:

  • Title showing scope (Global, Integration, or Action name)
  • Integration icon for context
  • Markdown editor for writing instructions
  • Cancel/Save buttons

Closing with unsaved changes prompts a confirmation dialog.

Instruction editor drawer
Instruction editor drawer

How Instructions Are Applied

When an action fails, the AI:

  1. Collects applicable instructions (global + integration + action)
  1. Combines them with error details and action context
  1. Generates explanation incorporating your guidance

Example:

“HubSpot Create or Update Contact" fails with these instructions loaded:

  • Global: "Our CRM uses UUID format for all IDs"
  • Integration: "We use 'external_id' property for matching"
  • Action: "Email field is required and must be lowercase"

Result: AI explanation includes all this context when analyzing the error.

FAQ

How many instructions can I create?

No limit.

Can I have multiple instructions at the same level?

One per scope: one global, one per integration, one per action.

Do instructions prevent errors?

No, they only improve error explanations. They don't affect action execution.

Who can manage instructions?

All members can view. Only Admins can create, edit, or delete.

Are instructions shared across workspaces?

No, they're scoped to each workspace.

Did this answer your question?
😞
😐
🤩

Last updated on January 21, 2026