New Standalone LLM Agent Paves the Way for Event-Driven AI

We are excited to announce the first step in our journey to build a comprehensive suite of event-driven AI capabilities for the enterprise: the beta release of our new Standalone LLM Agent. This new offering represents a significant milestone in our AI strategy, as we begin to develop our foundation for real-time, event-driven AI systems.

Why EDA Matters for AI

As enterprises begin to deploy AI at scale, they quickly discover a fundamental challenge: traditional integration approaches aren’t designed for the real-time, high-volume, asynchronous nature of modern AI workloads. While many organizations are experimenting with AI, scaling beyond proof-of-concept into production environments requires an architectural approach that can handle:

• Real-time updates and responses: AI systems need the latest information to provide accurate insights
• Back-pressure management: LLMs and vector databases can be easily overwhelmed without proper event throttling
• Reliable message delivery: Mission-critical AI applications can’t afford to lose inputs or responses
• Dynamic routing and filtering: Not every event needs AI processing – intelligent routing is essential for cost control

Event-driven architecture (EDA) is exactly the kind of foundation AI systems need. By decoupling producers and consumers through an event mesh, EDA lets you can create AI systems that are more resilient, scalable, and cost-effective than traditional request-response patterns allow.

Event-Driven AI Processing in the Cloud

Our new cloud-managed Standalone LLM Agent processes events from your event mesh, feeds them to large language models, and returns responses as new events enter the event mesh – all with the reliability, security, and governance you need.

Key capabilities of the beta release include:

• Configured and managed in the cloud
• Request/response processing with model providers with OpenAI-compatible APIs
• Configurable prompt
• Full event payload access for contextual processing
• Guaranteed delivery with configurable retry policies with your Solace broker
• No development, hosting or maintenance required

The agent integrates seamlessly with your existing event mesh, enabling you to add AI capabilities to your event flows without disrupting current operations. For example, a manufacturing system could use the Standalone LLM Agent to convert technical sensor alerts into actionable recommendations, or a customer service platform could translate incoming tickets into prioritized action items – all in real-time as events flow through your organization.

This approach also enables you to deploy powerful functionality in minutes with only a prompt. This agent can be thought of as an AI-layer microservice that you could create using coding, but now can be added without code and applied to real-time data in mere minutes. Models are becoming more intelligent, more reliable and less expensive all the time. And with smart topics you can control exactly which messages get sent to the LLM to reduce costs.

Built on Open Source

The Standalone LLM Agent is built on our open source solace-ai-connector project, which provides a Python framework for connecting Solace PubSub+ with popular AI services and tools. This open core approach allows for community-driven innovation while enabling us to provide enterprise-grade features for our customers.

If you’re a developer interested in exploring the technology behind our new agent, we encourage you to check out the repository, contribute, and provide feedback. Your input helps shape our product roadmap and ensures we’re meeting real-world needs.

Looking Ahead

The Standalone LLM Agent is just the beginning of our AI journey. In the coming months, we’ll be expanding our offerings to include:

1. Vector Database Micro-Integrations: Automatically prepare and store enterprise data for AI retrieval with support for popular vector databases like Qdrant and pgVector
2. RAG Query Agents: Enable efficient retrieval-augmented generation workflows by connecting LLMs with your enterprise knowledge base
3. Solace Agent Mesh: Orchestrate complex AI workflows with a complete suite of specialized agents working together through the event mesh

This evolution follows a clear progression from today’s event mesh to the integration of standalone AI agents and eventually to what we call Solace Agent Mesh – a framework that will enable autonomous agents to interact with your enterprise data and each other through a secure, reliable event infrastructure.

How to Get Started

The Standalone LLM Agent beta is available now for customers looking to begin their event-driven AI journey. Here are a few ideal starting use cases:

• Customer experience enhancement: Analyze incoming customer communications for sentiment, language, and intent, then route appropriately
• Content processing: Summarize, translate, or analyze documents as they flow through your organization
• Data enrichment: Add AI-generated insights, tags, or recommendations to your event streams
• Alert management: Transform technical alerts into actionable, human-readable recommendations

Specific examples are included at the end of this post. We also have a sentiment analysis feed available on solace.feeds.dev to help you try it out.

To get started, contact Solace support. We’re also looking for beta participants willing to provide detailed feedback as we refine the product for GA.

Demo

Conclusion

AI’s true potential in the enterprise will only be realized when it’s fully integrated with real-time data flows. With our Standalone LLM Agent and upcoming support of RAG workflows with by real-time event data, Solace is building the foundation that enterprises need to deploy AI at scale – reliable, secure, and designed for real-time operations from the ground up.

The future of enterprise AI isn’t just about smarter models – it’s about creating the infrastructure that allows those models to work together with your existing systems in a seamless, event-driven architecture. We’re excited to be at the forefront of this evolution and look forward to partnering with you on your AI journey.

Stay tuned for more updates as we continue to expand our agentic AI capabilities and move toward the full Agent Mesh vision!

Prompt Examples for Real-Time Data

Manufacturing and Operations

Anomaly Detection

Sample Event Data

{
  "timestamp": "2025-02-04T14:32:00Z",
  "equipment_id": "EQ-12345",
  "sensor_type": "temperature",
  "sensor_value": 105.5,
  "threshold": 100.0,
  "status": "alert"
}

Prompt

You are a manufacturing expert. Analyze this sensor data event—which includes metadata such as equipment ID, sensor type, timestamp, sensor value, and threshold—and provide clear, actionable maintenance recommendations in 2-3 sentences. Here are preferred thresholds: XXXXX Here is the sensor reading: {{text://input.payload}}

Sample Output

{
  "input_event": {
    "timestamp": "2025-02-04T14:32:00Z",
    "equipment_id": "EQ-12345",
    "sensor_type": "temperature",
    "sensor_value": 105.5,
    "threshold": 100.0,
    "status": "alert"
  },
  "output": {
    "recommendation": "The temperature sensor on equipment EQ-12345 is registering 105.5°F, which exceeds the threshold of 100.0°F. It is recommended to immediately inspect the cooling system and verify sensor calibration to avoid potential overheating and equipment failure."
  }
}

Customer Experience

Sentiment Analysis

Sample Event Data

{
  "customer_id": "CUST-9876",
  "interaction_timestamp": "2025-02-04T15:45:00Z",
  "channel": "email",
  "message": "I am extremely disappointed with the service. The wait times were unacceptable."
}

Prompt

Analyze the following customer interaction, including metadata such as customer_id, channel, and interaction timestamp, and provide a sentiment score from 1-5 (where 1 is very negative and 5 is very positive). Include a brief explanation. Text to analyze: {{text://input.payload}}

Output

{
  "input_event": {
    "customer_id": "CUST-9876",
    "interaction_timestamp": "2025-02-04T15:45:00Z",
    "channel": "email",
    "message": "I am extremely disappointed with the service. The wait times were unacceptable."
  },
  "output": {
    "sentiment_score": 1,
    "explanation": "The customer's email reflects significant disappointment and frustration with long wait times, indicating a very negative experience."
  }
}

Translation

Sample Event Data

{
  "customer_id": "CUST-1122",
  "interaction_timestamp": "2025-02-04T12:00:00Z",
  "channel": "chat",
  "message": "Bonjour, j'ai besoin d'aide pour ma commande."
}

Prompt

Translate the following customer message into idiomatic English, maintaining the original tone and intent. Note that the event data includes metadata such as customer_id, interaction timestamp, and channel. Text to analyze: {{text://input.payload}}

Output

{
  "input_event": {
    "customer_id": "CUST-1122",
    "interaction_timestamp": "2025-02-04T12:00:00Z",
    "channel": "chat",
    "message": "Bonjour, j'ai besoin d'aide pour ma commande."
  },
  "output": {
    "translation": "Hello, I need help with my order."
  }
}

Feedback Summarization

Sample Event Data

{
  "customer_id": "CUST-3344",
  "feedback_timestamp": "2025-02-04T11:30:00Z",
  "feedback": "The product quality was good, but shipping was delayed and customer service was unresponsive."
}

Prompt

Summarize the key points from this customer feedback in 3-4 bullet points, focusing on actionable insights. The event includes metadata such as customer_id and feedback_timestamp. Feedback: {{text://input.payload}}

Output

{
  "input_event": {
    "customer_id": "CUST-3344",
    "feedback_timestamp": "2025-02-04T11:30:00Z",
    "feedback": "The product quality was good, but shipping was delayed and customer service was unresponsive."
  },
  "output": {
    "summary_points": [
      "Product quality meets expectations.",
      "Shipping experienced significant delays.",
      "Customer service did not respond promptly."
    ]
  }
}

Multi-level Summarization

Sample Event Data

{
  "content_id": "DOC-7788",
  "submission_timestamp": "2025-02-04T16:20:00Z",
  "content": "In recent updates, our new software release includes several bug fixes, performance improvements, and user interface enhancements that will benefit both technical users and general audiences."
}

Prompt

Create two summaries of the following content: one technical summary for engineers and one general summary for a broad audience. The event includes metadata such as content_id and submission_timestamp. Content: {{text://input.payload}}

Output

{
  "input_event": {
    "content_id": "DOC-7788",
    "submission_timestamp": "2025-02-04T16:20:00Z",
    "content": "In recent updates, our new software release includes several bug fixes, performance improvements, and user interface enhancements that will benefit both technical users and general audiences."
  },
  "output": {
    "technical_summary": "The latest software release addresses multiple bugs, enhances performance metrics, and refines the UI to improve system stability and developer usability.",
    "general_summary": "Our new update fixes common issues, runs faster, and looks better for a smoother user experience."
  }
}

Financial Services

News Summarization Prompt

Sample Event Data

{
  "article_id": "NEWS-8899",
  "publication_date": "2025-02-04T08:00:00Z",
  "title": "Market Update: Federal Reserve Hints at Interest Rate Adjustment",
  "article_content": "In a recent press conference, the Federal Reserve signaled potential changes in interest rates amid inflation concerns. Market analysts predict volatility in the coming weeks."
}

Prompt

As a financial analyst, provide a concise 2-sentence summary of the key market implications from this news article. The event includes metadata such as article_id, title, and publication_date. Article: {{text://input.payload}}

Output

{
  "input_event": {
    "article_id": "NEWS-8899",
    "publication_date": "2025-02-04T08:00:00Z",
    "title": "Market Update: Federal Reserve Hints at Interest Rate Adjustment",
    "article_content": "In a recent press conference, the Federal Reserve signaled potential changes in interest rates amid inflation concerns. Market analysts predict volatility in the coming weeks."
  },
  "output": {
    "summary": "The Federal Reserve’s indication of possible interest rate adjustments is expected to introduce market volatility in the near term. Analysts suggest investors prepare for fluctuations as inflation concerns remain a key focus."
  }
}

Trading Data Narration Prompt

Sample Event Data

{
  "trade_id": "TRD-1020",
  "timestamp": "2025-02-04T10:15:00Z",
  "trading_metrics": {
    "volume": 500000,
    "average_price": 102.5,
    "daily_high": 105.0,
    "daily_low": 100.0
  }
}

Prompt

Convert these trading metrics into a natural language market insight suitable for a financial advisor’s client update. The event data includes metadata such as trade_id and timestamp. Metrics: {{text://input.payload}}

Output

{
  "input_event": {
    "trade_id": "TRD-1020",
    "timestamp": "2025-02-04T10:15:00Z",
    "trading_metrics": {
      "volume": 500000,
      "average_price": 102.5,
      "daily_high": 105.0,
      "daily_low": 100.0
    }
  },
  "output": {
    "narrative": "Today's trading activity saw a volume of 500,000 shares with an average price of $102.50. The market traded between $100.00 and $105.00, indicating moderate volatility that clients should monitor closely."
  }
}

Data Transformation

Format Conversion Prompt

Sample Event Data

{
  "record_id": "REC-3344",
  "data": "Name: Widget A; Price: 20.5; Stock: 150"
}

Prompt

Convert the following data into a well-structured JSON object, inferring appropriate field names where needed. The event includes metadata such as record_id. Data: {{text://input.payload}}

Output

{
  "input_event": {
    "record_id": "REC-3344",
    "data": "Name: Widget A; Price: 20.5; Stock: 150"
  },
  "output": {
    "record_id": "REC-3344",
    "product": {
      "name": "Widget A",
      "price": 20.5,
      "stock": 150
    }
  }
}

Data Normalization Prompt

Sample Event Data

{
  "product_id": "PROD-5566",
  "description": "widget a: 20 units per box, price in dollars: $15, dimensions: 10x5x3 inches"
}

Prompt

Standardize the following product description to match our company style guide: use metric units, capitalize product names, and format specifications consistently. The event includes metadata such as product_id. Description: {{text://input.payload}}

Output

{
  "input_event": {
    "product_id": "PROD-5566",
    "description": "widget a: 20 units per box, price in dollars: $15, dimensions: 10x5x3 inches"
  },
  "output": {
    "product_name": "Widget A",
    "packaging": "20 units per box",
    "price": "15 USD",
    "dimensions": "25.4 x 12.7 x 7.6 cm"
  }
}

QA & Compliance

Compliance Check

Sample Event Data

{
  "product_id": "MED-7788",
  "description": "Acme Pain Reliever 500mg - Use as directed. Contains aspirin. Not recommended for children under 12. Consult your doctor if symptoms persist."
}

Prompt

Review this product description for compliance with FDA labeling requirements. The event includes metadata such as product_id. Flag any potential issues and suggest corrections. Description: {{text://input.payload}}

Output

{
  "input_event": {
    "product_id": "MED-7788",
    "description": "Acme Pain Reliever 500mg - Use as directed. Contains aspirin. Not recommended for children under 12. Consult your doctor if symptoms persist."
  },
  "output": {
    "compliance_review": "The product description for Acme Pain Reliever 500mg appears to meet basic FDA guidelines; however, it should include detailed dosage instructions, potential side effects, and contraindications. Consider adding a warning section regarding allergic reactions to aspirin."
  }
}

Communication Review Prompt

Sample Event Data

{
  "communication_id": "COMM-9900",
  "timestamp": "2025-02-04T13:50:00Z",
  "message": "Dear client, please note that the investment carries inherent risks, and past performance is not indicative of future results. Please ensure all disclosures are accurate."
}

Prompt

Analyze this customer communication for potential compliance violations with financial services regulations. The event includes metadata such as communication_id and timestamp. Highlight any concerning language and suggest revisions. Message: {{text://input.payload}}

Output

{
  "input_event": {
    "communication_id": "COMM-9900",
    "timestamp": "2025-02-04T13:50:00Z",
    "message": "Dear client, please note that the investment carries inherent risks, and past performance is not indicative of future results. Please ensure all disclosures are accurate."
  },
  "output": {
    "compliance_review": "The communication is generally compliant; however, the phrase 'inherent risks' should be further clarified with specifics to avoid ambiguity. It is recommended to include a brief disclaimer outlining the exact risks and regulatory disclosures required."
  }
}