sigiro docs

sigiro is the detect and diagnose layer for software that repairs itself. OpenTelemetry goes in; ranked results and runnable SQL come out, for agents.

sigiro is an observability backend for agents, not for dashboards. It reads OpenTelemetry. It answers the two questions that an automatic loop must answer before the loop repairs a fault: what changed, and why.

These docs have four parts. Each part answers a different kind of question, so go to the part that matches the question you have now.

Tutorials

A lesson. Read this part first, because it assumes no sigiro knowledge. You follow each step, and you see a result at each step.

• Quickstart — run the server, send telemetry to it, and ask it what changed and why. This takes five minutes.

How-to guides

Directions for one task. Read this part when you know what you want, and you need the steps to get it.

• Instrument your code with an agent — give one prompt to an agent. The agent connects OpenTelemetry to your service and points it at sigiro.
• Send telemetry to hosted sigiro — run no server yourself. Send OTLP to a host that we operate, with an OAuth access token.
• Send your agent’s own telemetry to sigiro — point Claude Code, Codex, Hermes Agent, or Pi at sigiro.
• Debug and optimize an agent — find where a session spent time and tokens, remove wasted work, and compare the next run with evidence.
• Connect sigiro to Executor — import the OpenAPI API or the documentation MCP server into an agent gateway.
• Send Pyroscope profiles to sigiro — configure a verified profiling SDK, then read the profile rows back with SQL.

Reference

Facts about the machine. Read this part while you work, to confirm a name, a type or a limit.

• API reference — the HTTP operations currently described by the server’s /openapi.json. Ingest protocols that are not in that document, including the Pyroscope-compatible routes, are documented in the how-to guides.
• CLI reference — seven subcommands, every flag, and every environment variable they read.

An agent that reads /openapi.json needs no other information from us.

Explanation

Why sigiro works as it does. Read this part when you want to know the reasons, the costs and the alternatives. You do not need a terminal.

• About the tables, and which machine they describe — the stored signals and how to avoid answering with data from the wrong machine.
• About evidence instead of a dashboard — why the answer is a ranked list with a runnable query on every row.
• About deploy detection, and what your spans must carry — sigiro reads one attribute to find a deploy, and it says nothing when that attribute is absent.