Skip to content
Pitot

Observability your agents can operate.

Send OpenTelemetry to one binary. Investigate with Pitot's built-in agent, or connect yours over MCP.

Install PitotConnect your agent

v2026.11.0-alpha: 507,311 accepted items per second on average over 306.8 seconds: 61,571 spans, 224,112 log records and 221,628 metric points per second. Stacked areas show trailing five-second rates from Pitot's accepted-row counters, starting with the first full window. Measured mixed-sustained trial.

Sustained mixed ingest507,311telemetry items/sFive-minute mixed-ingest average
Spans 61,571Logs 224,112Metrics 221,628
507,311 items/s · 306.8 seconds of measured mixed ingestv2026.11.0-alpha: 507,311 accepted items per second on average over 306.8 seconds: 61,571 spans, 224,112 log records and 221,628 metric points per second. Stacked areas show trailing five-second rates from Pitot's accepted-row counters, starting with the first full window. Measured mixed-sustained trial.
Full measured run05:06 / 05:06

Replay of the measured run · Hetzner ccx33 · 8 vCPUHow we measured →

  • MCP for your agents
  • OTLP traces, logs and metrics
  • Parquet you own
  • Apache-2.0
Dashboards

Dashboards from a sentence.

Section titled “Dashboards from a sentence.”

Ask for the whole store’s request rate, errors and p95 latency by service, its dependency map, slowest operations, recent errors and slowest traces. The agent checks every panel, then saves the dashboard.

Pitot agent building and saving a store overview dashboard, then a tour of its panels.
Pitot agent building and saving a store overview dashboard, then a tour of its panels.

A one-sentence request becomes a saved eleven-panel dashboard: request rate, error rate, p95 latency and error log stats, by-service time series, a service dependency map, slowest operations, recent error logs and slowest traces.

Pitot interface · sample telemetryView full sizeView full size
Investigation

Ask what broke.

Section titled “Ask what broke.”

The agent queries traces, logs and metrics, then explains what the evidence shows.

Pitot agent answering which service is slowest, with telemetry evidence inline.
Pitot agent answering which service is slowest, with telemetry evidence inline.

The question "Which service is slowest right now?" leads to a system health check and a service latency table, then an answer that names accounting and explains its p95 latency.

Pitot interface · sample telemetryView full sizeView full size
MCP

Use the same tools from your agent.

Section titled “Use the same tools from your agent.”

Query telemetry, inspect traces and save dashboards through typed MCP tools.

  • get_telemetry_schema
  • query_telemetry
  • search_logs
  • inspect_trace
  • preview_panels
  • create_dashboard

Connect an agent over MCP →

Claude Code · replace the URL with your instance
claude mcp add --transport http pitot \
  https://pitot.example.com/mcp

Any remote MCP client works. You authorize it in your browser.

Install Pitot. Send OpenTelemetry.

One binary and one data directory. Add an AI provider key to enable the built-in agent.

curl -fsSL https://raw.githubusercontent.com/labstack/pitot/main/scripts/install.sh | sh
Read the install guideConnect your agent

Deployment requirements · AI configuration · Storage and retention