Free
$0per workspace / month30,000 ingestion units a month, 14-day retention, 1 project, 2 seats. No credit card. Official source 1 ↗ · Official source 2 ↗
Reviewed tool profile
telemetry.dev
OpenTelemetry-based tracing of LLM calls and AI agents, with token cost, latency, errors and alerts.
Sources checked
- Pricing
- Freemium, priced per workspace per month. Free is $0 with no card: 30,000 ingestion units a month (one per span, log record or metric point), 14-day retention, one project and two seats. Pro is $25 a month with 500,000 units, 30-day retention, 5 projects and 5 seats. Team is $50 a month with 2 million units, 90-day retention, unlimited projects and seats and priority support. Official source 1 ↗ · Official source 2 ↗
- Available as
- A web dashboard with first-party SDKs for TypeScript (Node, Vercel Edge and Cloudflare Workers) and Python; any other language can send standard OTLP over HTTP. · Integrations cover model providers such as OpenAI, Anthropic and the Claude SDK, frameworks such as the Vercel AI SDK, and coding agents such as Claude Code and Cursor. Official source 1 ↗ · Official source 2 ↗ Official source 1 ↗ · Official source 2 ↗
- Primary group
- Coding & Development
Also consider
See all 6 similar tools ↓- ColdteaControl layer for AI-written software: map every pull request, test the flows it touches with agents and watch it in production.Free plan
- Pluto DoorA Mac SSH client with Ori, an AI agent that investigates and fixes your servers after you approve.Free trial
- NitroStackPlatform for building, testing and hosting MCP servers: open-source TypeScript SDK, NitroStudio desktop app and NitroCloud hosting.Free plan
What it does
telemetry.dev is an LLM observability service for developers building AI applications and agents. Built on OpenTelemetry, it groups each request or agent run into one trace of model calls, tool steps and retrievals, with token usage, cost, latency and errors for each step. You connect it with its TypeScript or Python SDK, a provider or framework integration, or an existing OTLP exporter, and it is positioned against LangSmith and Langfuse.
Official source 1 ↗ · Official source 2 ↗Features
- Traces for agent runs. Each request or agent run becomes one trace of model calls, tool steps and retrievals, with the duration of each step, so you can find the slow step or the failed call. Official source 1 ↗
- Cost tracking. Cost is calculated at ingest from reported token usage and a model-pricing catalogue, with separate rates for cache reads and writes where available. Official source 1 ↗
- MCP analytics. Traces Model Context Protocol clients and servers in the same trace and sorts failures by error type and JSON-RPC code, so you can see which tools fail and which calls never reached a server. Official source 1 ↗
- Alerts against your own history. Every 5 minutes, alerts compare the project with its own history for tool errors, model errors, budget and agent loops, and link to the failed spans. They go to email, Slack, Discord or webhooks. Official source 1 ↗
- Prompt management and evals. Version and roll out prompts without a deploy and compare versions on real calls; eval runs are traced with their cost and errors. Official source 1 ↗
- Read-only MCP server for your agent. Connect Claude, ChatGPT, Cursor or another MCP client so an agent can search traces, read spans and summarise usage and cost. The server is read-only. Official source 1 ↗
- Setup prompt for coding agents. You can hand setup to your coding agent with one copied prompt instead of following the guide by hand. Official source 1 ↗
Pricing & free limits
Pro
$25per workspace / month500,000 ingestion units a month, 30-day retention, 5 projects, 5 seats. Ingestion continues above the allowance with no automatic overage charges. Official source 1 ↗ · Official source 2 ↗
Team
$50per workspace / month2 million ingestion units a month, 90-day retention, unlimited projects and seats, priority support. Official source 1 ↗ · Official source 2 ↗
Billing details
Polar handles checkout as merchant of record. Plans can be upgraded, downgraded or cancelled at any time, with no contracts. Official source 1 ↗ · Official source 2 ↗ · Official source 3 ↗
Paid plans are not cut off above their allowance, but the team contacts workspaces that keep exceeding it. Official source 1 ↗
A gift link can give a workspace Pro or Team for a set time without a card. Official source 1 ↗
Free access & limits
Free includes 30,000 ingestion units a month with 20% headroom; requests that would take the month past 36,000 units are rejected until the next calendar month or an upgrade. Official source 1 ↗
Free keeps data for 14 days, with one project and two seats. Official source 1 ↗
Best for & limitations
Find which tool call or model call made an agent run slow or fail, and what it cost. Official source 1 ↗
Get warned when an agent loops, a tool starts failing or monthly spend is on track to pass a budget. Official source 1 ↗
Ask your coding agent in the editor why the last run failed, using the read-only MCP server. Official source 1 ↗
- Prompts are captured by default. Prompt and response capture is on by default. Turning it off per environment does not stop transmission, and redaction patterns cannot catch all sensitive data, so sensitive content has to be removed before export. Official source 1 ↗
- Detail depends on your instrumentation. Trace detail depends on what your instrumentation emits, and cost is unavailable when model pricing or token usage is missing. Official source 1 ↗
- MCP tracing is TypeScript SDK v2 only. MCP analytics works with the official TypeScript MCP SDK v2. Official source 1 ↗
- US hosting, developers 16+. Servers and databases run in Northern Virginia, United States, and the service is not for anyone under 16. Official source 1 ↗
Similar tools
Tools like telemetry.dev
Published tools that share its categories.
ColdteaControl layer for AI-written software: map every pull request, test the flows it touches with agents and watch it in production.
Visit Pluto DoorA Mac SSH client with Ori, an AI agent that investigates and fixes your servers after you approve.
Visit NitroStackPlatform for building, testing and hosting MCP servers: open-source TypeScript SDK, NitroStudio desktop app and NitroCloud hosting.
Visit 
NavigeA one-line proxy that prices every AI agent call by business job, enforces policies and approvals, and cuts model spend with caching and routing.
Visit 
PolylaneAI SRE that watches your cloud and telemetry, finds root causes with cited evidence, reviews PRs against production and opens fix pull requests.
Visit 
TidenVerify that code from your AI coding agents does what you asked, with versioned specs, linked tests and a quality gate on every merge.
Visit