发送 OpenTelemetry，用 SQL 查询。

sigiro 存储你的链路、日志、指标和性能剖析数据。通过一个 HTTP API 用 SQL 查询：智能体走 CLI 和 MCP，人看仪表盘。

检测 · sigiro

$ curl -sS -X POST localhost:9999/v1/query --data-binary \
    "SELECT count(*) AS failed_spans FROM sigiro_spans
     WHERE service_name = 'checkout' AND status_code = 2
       AND timestamp > CURRENT_TIMESTAMP - INTERVAL '30 minutes'"
[{ "failed_spans": 246 }]

诊断 · sigiro

$ curl -sS -X POST localhost:9999/v1/query --data-binary \
    "SELECT span_name, count(*) AS failures FROM sigiro_spans
     WHERE service_name = 'checkout' AND status_code = 2
       AND timestamp > CURRENT_TIMESTAMP - INTERVAL '30 minutes'
     GROUP BY span_name ORDER BY failures DESC LIMIT 5"
[{ "span_name": "POST /api/orders", "failures": 246 }]

修复 · 你的智能体

$ claude "checkout times out since 9f2c1ab"
read   src/orders/upstream.ts
edit   src/orders/upstream.ts   +4 −1
test   42 passed
push   fix: bound the orders upstream call at 5s

验证 · 你的流水线，用我们的证据

$ curl -sS -X POST localhost:9999/v1/query --data-binary \
    "SELECT count(*) AS errors_last_5m FROM sigiro_spans
     WHERE service_name = 'checkout' AND status_code = 2
       AND timestamp > CURRENT_TIMESTAMP - INTERVAL '5 minutes'"
[{ "errors_last_5m": 0 }]

沉淀 · 你的记忆库

$ cat playbooks/checkout-upstream-timeout.md
## checkout · upstream timeout

observed   246 failed checkout spans in a 30-minute window
cause      unbounded upstream call, shipped in 9f2c1ab
fix        5s deadline on the orders client
watch      the same shape on any service calling orders

factory loop · five stages

部署在你自己的服务器上就免费，sigiro 不会把任何数据发给我们。注册则使用托管测试版，遥测数据存放在我们运行的服务器上。

接入

sigiro 使用标准 OTLP，监听标准端口。

你的终端

$ curl -fsSL https://sigiro.com/install | sh
installed ~/.local/bin/sigiro

$ sigiro serve &
[1] 41802
listening on 4317, 4318, 9999

$ export OTEL_EXPORTER_OTLP_ENDPOINT=http://localhost:4318
$ export OTEL_SERVICE_NAME=checkout
$ npm start &
[1] 41823
checkout listening on :3000

$ sigiro status
ready · OTLP 4317 and 4318 · service checkout · 18,402 spans in 5 min

提问

通过 HTTP 发送 SQL，答案以 JSON 行的形式返回。

你的终端

$ curl -sS -X POST localhost:9999/v1/query --data-binary \
    "SELECT span_name, count(*) AS calls FROM sigiro_spans
     WHERE service_name = 'checkout'
       AND timestamp > CURRENT_TIMESTAMP - INTERVAL '30 minutes'
     GROUP BY span_name ORDER BY calls DESC LIMIT 5"
[
  { "span_name": "POST /api/orders", "calls": 1284 },
  { "span_name": "GET /api/cart", "calls": 342 }
]

关联

智能体就是一个会调用模型的服务。同一条链路里两者都在。

你的终端

$ cat models.sql
SELECT s.span_attributes ->> 'gen_ai.request.model' AS model,
       count(DISTINCT s.span_id) AS calls, max(s.duration) AS slowest
FROM sigiro_spans s JOIN sigiro_logs l USING (trace_id)
WHERE s.service_name = 'checkout' AND l.severity_number >= 17
  AND s.timestamp > now() - INTERVAL '30 minutes'
GROUP BY model ORDER BY slowest DESC

$ curl -sS -X POST localhost:9999/v1/query \
    --data-binary "$(cat models.sql)"
MODEL           CALLS  SLOWEST
gpt-5.2-mini      246    30.0s
claude-haiku-4.6   31     2.1s

深入排查

列出失败的请求。每一行都带有 trace_id，可以从 /v1/traces 取回完整链路。

你的终端

$ curl -sS -X POST localhost:9999/v1/query --data-binary \
    "SELECT trace_id, duration, status_message FROM sigiro_spans
     WHERE service_name = 'checkout' AND status_code = 2
       AND timestamp > CURRENT_TIMESTAMP - INTERVAL '30 minutes'
     ORDER BY timestamp DESC LIMIT 3"
[
  { "trace_id": "9f2c1ab4e77d", "duration": 30012,
    "status_message": "context deadline exceeded" },
  { "trace_id": "3b81fe0a2c56", "duration": 30004,
    "status_message": "context deadline exceeded" }
]

怎么用

四个步骤。你依次完成接入、提问、关联和深入排查。

把 OpenTelemetry 导出器指向 sigiro，然后用 SQL 提问。下面每一步都是一次终端会话。

接入1 of 4

$ curl -fsSL https://sigiro.com/install | sh

installed ~/.local/bin/sigiro

$ sigiro serve &

[1] 41802

listening on 4317, 4318, 9999

$ export OTEL_EXPORTER_OTLP_ENDPOINT=http://localhost:4318

$ export OTEL_SERVICE_NAME=checkout

$ npm start &

[1] 41823

checkout listening on :3000

$ sigiro status

ready · OTLP 4317 and 4318 · service checkout · 18,402 spans in 5 min

sigiro 使用标准 OTLP，监听标准端口。

$ curl -fsSL https://sigiro.com/install | sh
installed ~/.local/bin/sigiro

$ sigiro serve &
[1] 41802
listening on 4317, 4318, 9999

$ export OTEL_EXPORTER_OTLP_ENDPOINT=http://localhost:4318
$ export OTEL_SERVICE_NAME=checkout
$ npm start &
[1] 41823
checkout listening on :3000

$ sigiro status
ready · OTLP 4317 and 4318 · service checkout · 18,402 spans in 5 min

$ curl -sS -X POST localhost:9999/v1/query --data-binary \
    "SELECT span_name, count(*) AS calls FROM sigiro_spans
     WHERE service_name = 'checkout'
       AND timestamp > CURRENT_TIMESTAMP - INTERVAL '30 minutes'
     GROUP BY span_name ORDER BY calls DESC LIMIT 5"
[
  { "span_name": "POST /api/orders", "calls": 1284 },
  { "span_name": "GET /api/cart", "calls": 342 }
]

$ cat models.sql
SELECT s.span_attributes ->> 'gen_ai.request.model' AS model,
       count(DISTINCT s.span_id) AS calls, max(s.duration) AS slowest
FROM sigiro_spans s JOIN sigiro_logs l USING (trace_id)
WHERE s.service_name = 'checkout' AND l.severity_number >= 17
  AND s.timestamp > now() - INTERVAL '30 minutes'
GROUP BY model ORDER BY slowest DESC

$ curl -sS -X POST localhost:9999/v1/query \
    --data-binary "$(cat models.sql)"
MODEL           CALLS  SLOWEST
gpt-5.2-mini      246    30.0s
claude-haiku-4.6   31     2.1s

$ curl -sS -X POST localhost:9999/v1/query --data-binary \
    "SELECT trace_id, duration, status_message FROM sigiro_spans
     WHERE service_name = 'checkout' AND status_code = 2
       AND timestamp > CURRENT_TIMESTAMP - INTERVAL '30 minutes'
     ORDER BY timestamp DESC LIMIT 3"
[
  { "trace_id": "9f2c1ab4e77d", "duration": 30012,
    "status_message": "context deadline exceeded" },
  { "trace_id": "3b81fe0a2c56", "duration": 30004,
    "status_message": "context deadline exceeded" }
]

Start

install the CLI

curl -fsSL https://sigiro.com/install | sh

Installs a single binary to ~/.local/bin/sigiro. Reads the script first if you would rather: sigiro.com/install.

or give this to your agent

Install and run sigiro, the observability data layer, on this machine. Run: curl -fsSL https://sigiro.com/install | sh — then start it with: sigiro serve. It listens on 4317 and 4318 for OTLP and on 9999 for its HTTP API. Point this project’s OpenTelemetry exporter at http://localhost:4318. Then register `sigiro mcp` as a stdio MCP server (for Claude Code: claude mcp add sigiro -- sigiro mcp) so you can query the telemetry directly with its list_services and run_query tools.

Paste into Claude Code, Codex, Cursor or any MCP-capable agent. It installs sigiro, starts it, and registers sigiro mcp as an MCP server.

自动发现。你的智能体不用人帮忙就能读懂 API 描述。

一份公开文档描述了全部接口。

GET /openapi.json · 公开，无需密钥

[
  "/v1/query",
  "/v1/services",
  "/v1/traces/{trace_id}"
]

你能得到什么

不用迁移。 sigiro 能对接你正在用的工具。

从这些地方发送遥测数据

OpenTelemetry
Node.js
Python
Go
Java
.NET
Ruby
Rust
Claude Code
Codex
Vercel AI SDK
LangChain
Ollama

在这些平台上运行

Docker
Kubernetes
Fly.io
Railway
Render
Cloudflare

这些工具可以读取

Claude Code
Cursor
Codex
任何会说 SQL 的工具

答案可以重跑: 每个答案都是一条 SQL 查询。你、你的智能体或同事都可以再跑一遍。
所有信号在一处: 链路、日志、指标和性能剖析都是表，一条查询就能关联起来。
五分钟装好: 一条命令，两个环境变量。代码一行都不用改。
自己部署就免费: 不按主机、席位或数据量收费。托管测试版尚未公布价格，欢迎来聊。
数据留在你手里: 你把 sigiro 装在自己的服务器上。sigiro 不会把任何数据发给我们。托管测试版例外：遥测数据存放在我们这里。
sigiro 监控 sigiro: sigiro 把自己的遥测数据发给 sigiro。我们就是这样发现自己的故障的。

发送 OpenTelemetry，用 SQL 查询。

检测 · sigiro

诊断 · sigiro

修复 · 你的智能体

验证 · 你的流水线，用我们的证据

沉淀 · 你的记忆库

四个步骤。 你依次完成接入、提问、关联和深入排查。

Start

自动发现。 你的智能体不用人帮忙就能读懂 API 描述。

不用迁移。 sigiro 能对接你正在用的工具。

四个步骤。你依次完成接入、提问、关联和深入排查。

自动发现。你的智能体不用人帮忙就能读懂 API 描述。