AI Assistant

Ask questions about your telemetry in natural language and let the built-in agent investigate for you. It reasons step by step and calls read-only tools across your APM traces, RUM sessions, logs, errors, uptime checks, infrastructure, databases, alerts, and saved views — then returns a grounded answer with every reasoning step and tool call kept visible and auditable. Choose your execution engine: run a model fully in your browser via WebGPU for total privacy, bring your own cloud API key, or point at any self-hosted OpenAI-compatible endpoint. Conversation history is optional — you can keep chats entirely off the server.

Prerequisites

  • An active Senzor workspace with telemetry to investigate.
  • For local inference: a WebGPU-capable browser (Chrome 113+ or Edge 113+) with a compatible GPU.
  • For cloud providers: your own API key for the chosen provider. Keys are stored locally in your browser and calls go directly to the provider — they are never sent to or stored on Senzor servers.
  • For a custom endpoint: a reachable OpenAI-compatible API URL (e.g., Ollama, vLLM, or LM Studio).

1. Dashboard Configuration

Before injecting code into your application, you must provision an API key from the Senzor dashboard.

1

Open the Assistant

Navigate to AI Assistant in the dashboard sidebar.

2

Choose an Execution Engine

Select Local (WebLLM) to run in-browser via WebGPU, a cloud provider (OpenAI, Anthropic, Google AI, Groq, Mistral, or OpenRouter) with your own key, or a Custom Endpoint for any OpenAI-compatible API.

3

Configure Access

For local inference, pick a model sized to your GPU's available VRAM and load it. For a cloud provider, paste your API key. For a custom endpoint, enter the endpoint URL and model name.

4

Start Investigating

Ask a question in plain English (e.g., 'Why did checkout latency spike at 14:30?') and watch the agent reason and query your telemetry to a grounded answer.

Troubleshooting & Edge Cases

WebGPU is not available in this browser.

Local (WebLLM) inference requires WebGPU. Use Chrome 113+ or Edge 113+ with a compatible GPU, or switch to a cloud provider or custom endpoint, neither of which needs WebGPU.

The local model fails to load or runs very slowly.

The selected model likely needs more VRAM than your GPU reports. Choose a smaller model — for example Llama 3.2 3B or Qwen 2.5 1.5B — that fits within your hardware's available allocation.

Are my API keys or conversations sent to Senzor?

API keys are stored locally in your browser and cloud calls are made directly to the provider — keys never reach Senzor servers. Conversation history is optional; disable it to keep chats entirely off the server.