Back home

telemetry.dev

x 0.62
Visit source

Intro

telemetry.dev is an LLM observability platform for AI applications and agents, built on OpenTelemetry. It launched on Product Hunt on October 31, 2026, finishing #3 Product of the Day with about 128 upvotes; maker Ephraim Duncan promoted the launch on X.

The product shows each run of an AI agent as a single trace. Every model call, tool call and retrieval becomes a span with its duration, token usage, cost and error, so when an agent breaks in production you can see which step failed and what the run cost. The launch added prompt management, evals, alerts, a CLI and an MCP server.

Explain

Key features

  • Traces and spans: one trace per request or agent run; model calls, tool steps and retrievals are spans with latency, tokens, cost and errors.
  • Prompt management: change prompts without redeploying.
  • Evals: run a new prompt version on a dataset and compare scores with production traces before going live.
  • Alerts: notifications when errors pile up, costs spike, or an agent gets stuck in a loop.
  • CLI + MCP server: your coding agent (Claude Code, Codex, etc.) can read your traces directly.
  • Integrations: TypeScript and Python SDKs for OpenAI, Anthropic, the Vercel AI SDK, LangChain and more, or plug in an existing OpenTelemetry exporter.

Pricing

A free plan includes 30,000 ingestion units per month (one per span, log record or metric data point) with no credit card; paid Pro plans add more volume.

How it compares

It competes with tools like Langfuse, LangSmith, Helicone and Arize Phoenix. Its pitch is being OpenTelemetry-native and agent-focused, with loop and cost alerts plus an MCP server for agent self-debugging.

Usage

Getting started

  1. Sign up at https://telemetry.dev/ (free plan, no card).
  2. Install the TypeScript or Python SDK, or point your existing OpenTelemetry exporter at telemetry.dev.
  3. Run your agent; each run appears as a trace with per-span tokens, cost and errors.
  4. Move prompts into prompt management, build a small dataset, and run evals before shipping prompt changes.
  5. Set alerts for error spikes, cost spikes and agent loops.
  6. Optionally connect the MCP server so your coding agent can query traces while debugging.

Good fits

  • Teams running LLM agents in production who need cost and failure visibility.
  • Developers already using OpenTelemetry who don't want another proprietary SDK.

Links