Skip to main content
Use this page to get one real tool invocation or AI interaction visible in the Agnost AI dashboard. Plan for about five minutes. The Agnost AI skill inspects your existing telemetry, chooses the smallest supported integration, and verifies the application path you actually run.

1. Get your organization ID

  1. Sign in at app.agnost.ai.
  2. Open Settings > Organization. Copy your organization ID: it’s a UUID.
In these docs, org_id and AGNOST_ORG_ID both mean your organization ID. It routes telemetry to your workspace but does not grant dashboard read access.

2. Give this prompt to your coding agent

From the root of the application you want to monitor, give your coding agent this single prompt. Replace your-org-id with the value from step 1.
Setup prompt
The skill asks only for facts it cannot discover, such as the restart command or an ambiguous application entrypoint. It does not replace working OpenTelemetry with a second integration or declare success from a standalone test event.
The npx command requires Node.js 18 or later. If your application does not use Node.js, the command still installs the agent skill; it does not add Node.js to your application runtime.

3. Run one real interaction

After the agent finishes its edits:
  1. Restart the application with its normal development command.
  2. Trigger one real chat, agent action, or MCP tool call through the application.
  3. Keep the Agnost onboarding screen open. It checks for incoming data automatically.
Do not use a synthetic ingestion request as the final verification. The goal is to prove that the real code path a user exercises produces useful telemetry.

Setup guides

Continue in the Get started sidebar and select the framework or application type you already use. Each AI framework guide keeps all three approaches together:
  • Agnost AI skill — recommended; it inspects, instruments, and verifies your app.
  • Agnost AI SDK — manually wrap the framework call you want to analyze.
  • OpenTelemetry — connect existing framework telemetry to Agnost AI.
Use one approach for a given call path at first. Enabling the SDK and OpenTelemetry around the same call can create duplicate interactions.

Agent frameworks

Set up Agnost AI for Agno, Anthropic, CrewAI, DSPy, LangChain, Mastra, OpenAI, and more.

Custom applications

Track custom Python or TypeScript AI interactions, or connect an existing OTel pipeline.

MCP servers

Track tool, resource, and prompt calls from a dedicated MCP server.

Platforms

Set up MCP Toolbox or Apify telemetry.

Verify your first useful trace

When the onboarding screen confirms the first event, choose Inspect first event. Verify these fields before sending more traffic:
  • The input and output match the interaction you just ran.
  • The user ID remains stable for the same user.
  • The conversation or session ID groups related turns together.
  • The success or error state is accurate.
  • The agent, model, and tool names identify the real execution path.
  • Tool spans appear under the turn that invoked them when your framework emits hierarchy.
Use Events as the unfiltered source of truth. SDK-tracked interactions also appear under Users, while MCP calls appear under Tools. Events usually arrive within a few seconds.

Nothing appeared

Check these in order:
  1. The application was restarted after environment or startup changes.
  2. The configured organization ID matches the current Agnost workspace.
  3. The interaction exercised the instrumented entrypoint rather than a test or fallback path.
  4. Application logs do not show exporter, network, or package errors.
  5. Only one integration wraps the call path; duplicate SDK and OpenTelemetry setup can create confusing results.

Turn the first trace into a failure detector

One event proves transport, not monitoring value. Complete the activation loop:
  1. Open Violations and define one expected behavior that can be judged from a conversation.
  2. Run a representative interaction that should pass or fail that rule.
  3. Open the supporting conversation before accepting the finding.
  4. When the same issue recurs, review its recommendation in Auto-improve and inspect any proposed code change before shipping it.
For example, a support agent rule might be: “Confirm identity before exposing account information.” Keep authorization enforcement in application code; the violation is monitoring evidence, not a security boundary.

Next steps

  • Conversations: review the complete interaction you sent under Users.
  • Events: inspect the underlying telemetry and verify its fields.
  • Violations: define expected behavior and review failures.
  • Improvements: understand how evidence becomes a proposed fix.
⌂ 换网址