OpenRouter AI

Overview

What it does: Integrates any AI model from the OpenRouter.ai library directly into your HubSpot workflows, enabling AI-powered content generation, analysis, and decision-making.

When to use it: When you need AI assistance for content creation, data analysis, lead scoring, email personalization, or any task that benefits from advanced language models like GPT-4, Claude, or specialized AI models.

Business value: Automates complex reasoning tasks, generates personalized content at scale, provides intelligent insights, and enables sophisticated AI-driven workflows without requiring separate AI integrations.

---

Quick Setup

Prerequisites

• OpenRouter.ai API account and API key
• API key configured in your Daeda Essentials app settings
• Understanding of AI prompting best practices
• Clear use case for AI integration in your workflow

Basic Configuration

1. Set up your OpenRouter.ai API key in the app settings
2. Add the “OpenRouter AI” action to your workflow
3. Select an AI model from the available options
4. Write your prompt with any HubSpot property tokens
5. Test with a few records to validate responses

---

Input Fields

Required Fields

AI Model

• Type: Dynamic Dropdown (populated from OpenRouter.ai)
• Description: The specific AI model to use for processing your prompt
• Example: gpt-5, claude-4.5-sonnet, llama-2-70b-chat
• Notes: Options include GPT models, Claude, Llama, and many specialized models

Prompt

• Type: Text Area
• Description: The instruction or question you want the AI to process
• Example: Write a personalized follow-up email for {{contact.firstname}} who downloaded our {{recent_content_title}} guide
• Notes: Can include HubSpot property tokens and complex instructions

Optional Fields

Provider

• Type: Dynamic Dropdown (populated based on selected model)
• Description: Specific AI provider to use when multiple providers offer the same model
• Default: Default (automatically balances across providers)
• Example: OpenAI, Anthropic, Together AI
• Notes: Useful for consistency or when you prefer specific providers

Web Search

• Type: Boolean Checkbox
• Description: Enables the AI to search the web for current information before responding
• Default: false
• Cost: Additional $0.10 USD per request
• Example: true for current events, market data, or recent information
• Notes: Provides more accurate and up-to-date responses

---

Output Fields (Action Outputs)

Response Fields

Response

• Type: Text
• Description: The AI’s response in plain text format
• Use: Store in HubSpot properties, use in subsequent workflow actions
• Example: Raw text response from the AI model

Response HTML

• Type: Text (HTML formatted)
• Description: The AI’s response formatted as HTML (useful for rich content)
• Use: Email templates, rich text properties, web content
• Example: Formatted response with headings, lists, and styling

Citations

• Type: Text
• Description: Source citations when web search is enabled
• Use: Track information sources, validate AI responses
• Example: wikipedia.org, company-website.com or “No citations”

---

Common Use Cases

Use Case 1: Personalized Email Generation

Scenario: Generate custom follow-up emails based on contact behavior and properties

Setup:

• Model: gpt-4
• Prompt: Write a personalized follow-up email for {{contact.firstname}} who works at {{company.name}} in the {{contact.industry}} industry. They recently {{recent_activity}}. Keep it professional and under 150 words.
• Web Search: false

Result: Personalized emails for each contact based on their specific context

Use Case 2: Lead Scoring Analysis

Scenario: Use AI to analyze lead quality based on multiple data points

Setup:

• Model: claude-3-sonnet
• Prompt: Analyze this lead: Company: {{company.name}}, Industry: {{company.industry}}, Size: {{company.numberofemployees}}, Recent Activity: {{recent_page_views}}. Provide a lead score from 1-10 and explain why.
• Web Search: false

Result: AI-generated lead scores with reasoning for sales team prioritization

Use Case 3: Content Summarization

Scenario: Summarize long-form content or meeting notes for quick review

Setup:

• Model: gpt-4
• Prompt: Summarize the following meeting notes in 3 bullet points: {{meeting.notes}}
• Web Search: false

Result: Concise summaries for easy consumption and follow-up

Use Case 4: Market Research Integration

Scenario: Get current market insights for deals in specific industries

Setup:

• Model: gpt-4
• Prompt: What are the current market trends and challenges in the {{company.industry}} industry that might affect a {{deal.dealtype}} deal worth {{deal.amount}}?
• Web Search: true

Result: Current market context to inform sales strategies

---

Advanced Configuration

Model Selection Strategy

Choose models based on your specific needs:

• GPT-4: Best for complex reasoning, creative tasks, and general intelligence
• Claude: Excellent for analysis, writing, and following detailed instructions
• Llama: Cost-effective for simpler tasks and high-volume processing
• Specialized Models: Task-specific models for coding, math, or domain expertise

Prompt Engineering Best Practices

• Be Specific: Clear, detailed instructions produce better results
• Use Examples: Show the AI what you want with examples
• Set Constraints: Specify length, format, tone, and style requirements
• Include Context: Provide relevant background information
• Test Iteratively: Refine prompts based on actual results

Cost Management

• Model Costs: Different models have different pricing (check OpenRouter.ai)
• Web Search: Adds $0.10 per request but provides current information
• Token Usage: Longer prompts and responses cost more
• Provider Selection: Some providers may be more cost-effective

API Behavior Details

• Request Timeout: Requests timeout after 60 seconds if the AI model doesn’t respond
• Retry Logic: Failed requests are automatically retried once before failing the workflow
• Rate Limiting: OpenRouter.ai handles rate limiting - workflows will wait and retry if limits are hit
• Token Limits: Each model has maximum token limits for input + output combined
• Response Streaming: Large responses are streamed back to prevent timeouts
• Error Propagation: API errors from OpenRouter.ai are passed through to help with debugging

---

How It Works

1. Model Selection: Choose from 100+ AI models available on OpenRouter.ai
2. Prompt Processing: Your prompt is processed with HubSpot property values
3. API Request: Secure request sent to OpenRouter.ai with your API key
4. AI Processing: Selected model processes your prompt (with web search if enabled)
5. Response Formatting: AI response converted to both text and HTML formats
6. Output Generation: Results available for use in subsequent workflow actions

---

Troubleshooting

Common Issues

Issue: “App account isn’t available” error

Symptoms: Workflow fails immediately with account error Solution: Ensure your OpenRouter.ai API key is properly configured in app settings

Issue: AI responses are inconsistent or poor quality

Symptoms: Responses don’t match expectations or vary widely Solution: Refine your prompt with more specific instructions and examples

Issue: Responses are cut off or incomplete

Symptoms: AI responses end abruptly or seem truncated Solution:

• The model may have hit its token limit - try a model with higher limits or shorter prompts
• Some models have built-in response length limits - specify desired length in your prompt
• Check if the response is actually complete but formatted unexpectedly
• HubSpot properties also have a character limit that you may be hitting

Issue: High costs from AI usage

Symptoms: Unexpected charges from OpenRouter.ai Solution: Monitor usage, choose cost-effective models, and avoid web search when not needed

---