Langfuse integration for end-to-end AI assistant tracing

Overview
Telnyx AI Assistants now send traces directly to Langfuse, giving full visibility into each conversation turn. Developers can inspect prompts, outputs, tool calls, latency, and errors in one place to debug faster and improve quality.

What's new

• Native Langfuse support: Send traces from assistants without custom instrumentation.
• End-to-end tracing: Capture LLM inputs, outputs, and conversation context.
• Tool execution logs: Record tool names, arguments, and responses.
• Deterministic trace grouping: Group events by conversation_id.
• Observability settings API: Enable or disable tracing via observability_settings.

Why it matters

• Reduces debugging time by exposing full request and response chains.
• Improves quality by reviewing real assistant behavior across turns.
• Tracks token usage and latency for cost and performance analysis.
• Simplifies root cause analysis for failed tool calls or responses.
• Enables consistent monitoring across multiple assistants or environments.

Example use cases

• Debug incorrect assistant responses by inspecting prompt history.
• Analyze latency across LLM calls and external tool executions.
• Monitor token usage trends for cost control.
• Review webhook tool behavior in production workflows.

Getting started

1. Create a Langfuse project and generate public and secret API keys.
2. Store keys as integration secrets in your Telnyx account.
3. Configure your assistant with observability_settings and set status to enabled.
4. Send requests through your assistant and view traces in Langfuse.
5. Use trace data to inspect prompts, outputs, and tool calls.

Learn more in our developer documentation or contact your Telnyx team.