text stringlengths 0 59.1k |
|---|
You can plug dozens of providers from OpenAI to Anthropic, Google to Groq with a single unified API. This _seriously_ speeds up development. |
But check out the problem. |
"I'm building awesome agents with Vercel AI SDK but I have _no clue what's happening in production_". |
Sound familiar? |
Building AI agents is easier with frameworks. But then comes the _production nightmare_: You don't know what tools your agent is executing, you can't see conversation flows, you can't monitor performance. |
When we're building VoltAgent, I kept asking myself "What's my agent doing now?" Traditional monitoring tools? Not so effective for AI applications. |
So we built the [VoltOps LLM Observability Platform](https://voltagent.dev/voltops-llm-observability/) which provides comprehensive monitoring for AI applications built with frameworks like Vercel AI SDK. |
:::tip Key Solution |
Whether you're building chatbots, content generators, or complex multi-agent systems with Vercel AI, you need visibility into what's happening under the hood. Our platform tracks everything from individual LLM calls to complex agent workflows, giving you the insights you need to optimize performance and debug issues in... |
::: |
## What Makes VoltOps LLM Observability Different |
Unlike traditional monitoring tools that focus on model metrics, VoltOps LLM Observability is designed specifically for **agent workflows**. When you're building with Vercel AI SDK, you need more than just token counts and response times. Here's what makes it unique: |
### Agent-Centric Approach |
VoltOps shows you what your **agent is doing**, not just how your model is performing: |
- **Conversation flows** - See the complete dialogue thread across multiple interactions. Track how users engage with your AI and where conversations succeed or fail. Unlike traditional tools that show isolated API calls, you get the full context of user journeys. |
- **Tool usage patterns** - Which tools are called when and why, with complete input/output tracking. See if your weather tool is being called appropriately, or if your database query tool is receiving the right parameters. This is crucial for debugging agent behavior in production. |
- **Multi-agent interactions** - How different agents collaborate and hand off tasks to each other. When you have a planning agent that delegates to execution agents, VoltOps tracks the entire workflow hierarchy, showing you parent-child relationships and data flow between agents. |
- **Decision tracking** - Why did the agent make that choice? See the reasoning chain that led to specific tool calls or responses. This helps you understand if your agent is making logical decisions or if there are gaps in your prompts or tool descriptions. |
 |
### Live Visualization & Immediate Debugging |
**Real-time insights while your agent runs:** |
- **Watch execution flow as it happens** - No waiting for batch processing or delayed dashboards. See your Vercel AI SDK agents execute in real-time, which is essential for debugging production issues as they occur. |
- **Spot problems instantly** - Instead of discovering failures hours later through logs, get immediate alerts when tools fail, when agents get stuck in loops, or when response times spike. This immediate feedback loop dramatically reduces time to resolution. |
- **Zero latency monitoring** - Between agent action and console visualization, there's virtually no delay. This real-time capability is particularly important for customer-facing applications where quick issue resolution directly impacts user experience. |
 |
### What You Actually See |
VoltOps visualizes the agent-specific data that matters for Vercel AI SDK applications: |
🔀 **Multi-agent coordination** - Parent-child relationships and hierarchies visualized as interactive flowcharts. See how your main conversation agent delegates to specialized tools like code generators, data analysts, or customer service bots. |
🛠️ **Tool execution flows** - Complete tool call sequences with inputs/outputs, execution times, and success rates. Know immediately if your database tool is receiving malformed queries or if your API integration is timing out. |
💬 **Conversation threading** - How messages connect across interactions, showing user intent progression and agent response patterns. This helps identify where users get confused or where your agent provides incomplete responses. |
🧠 **Agent decision making** - The reasoning behind each step, including which tools were considered but not used, and why certain responses were generated. This level of insight is impossible with traditional APM tools. |
This is fundamentally different from traditional LLM monitoring that focuses on token counts, response times, and model accuracy. VoltOps shows you the _behavior_ of your intelligent system, which is what actually matters when you're building production AI applications with Vercel AI SDK. |
## 1-Minute Setup: No Kidding |
Now let me demonstrate how to set this up for Vercel AI SDK. |
### Installation |
```bash |
npm install @voltagent/vercel-ai-exporter @opentelemetry/sdk-node @opentelemetry/auto-instrumentations-node |
``` |
Just 3 packages. That's all. |
### API Keys |
Go to [console.voltagent.dev](https://console.voltagent.dev), create an organization, create a project, get your keys: |
- `VOLTAGENT_PUBLIC_KEY` |
- `VOLTAGENT_SECRET_KEY` |
:::warning Security Note |
Keep your secret key safe! Never commit it to your repository. Always use environment variables in production. |
::: |
### OpenTelemetry Setup |
In your main file, do this: |
```typescript |
import { VoltAgentExporter } from "@voltagent/vercel-ai-exporter"; |
import { NodeSDK } from "@opentelemetry/sdk-node"; |
import { getNodeAutoInstrumentations } from "@opentelemetry/auto-instrumentations-node"; |
// Create VoltAgent exporter |
const voltAgentExporter = new VoltAgentExporter({ |
publicKey: process.env.VOLTAGENT_PUBLIC_KEY, |
secretKey: process.env.VOLTAGENT_SECRET_KEY, |
baseUrl: "https://api.voltagent.dev", // default |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.