sigiro is an observability backend for agents. You send OpenTelemetry to sigiro. You ask a question in SQL or with one investigate call. sigiro gives you a diagnosis, not a dashboard.

Does sigiro replace Datadog or Grafana?

sigiro replaces them when a machine reads the data. If a person must look at a chart, keep the dashboard. sigiro is for the steps after that.

Must I change my code?

No. sigiro uses standard OTLP on ports 4317 and 4318. Your OpenTelemetry SDK or collector sends to sigiro without a change.

Where does my data go?

You install sigiro on your servers. sigiro writes to your storage and sends no data to us. You do not make an account. The hosted service is different: it keeps your telemetry for you.

No. A page interrupts a person who cannot watch continuously. sigiro gives a table that a program reads in a loop.

How does sigiro know what to report without thresholds?

sigiro compares each service with its own recent history. A service that is always slow but steady reports nothing, because nothing changed. A service that becomes two times slower reports, because no rule was necessary.

Does sigiro sample the data?

No. sigiro keeps each request. Sample rates were necessary when no one could read all of the data. A machine can read all of the data.

Which query language must I learn?

None. You ask in SQL. Your agent writes SQL correctly. Each answer gives you the query that made it.

Can my agent use sigiro without me?

Yes. The agent reads the API description at /openapi.json. It needs no key and no tool definition. Then it asks in SQL.

What does sigiro cost?

sigiro is free when you install it, and the storage is yours. There are no fees for hosts, seats or data. The hosted service costs approximately $14 each month. Access is by invitation.

What does sigiro not do?

sigiro does not write your patches, run your pipeline, or evaluate alert rules. sigiro has no dashboard builder and no full-text log index. Questions are structured, because that is how an agent asks.

The data layer for software that repairs itself.

Send OpenTelemetry to sigiro. When a service fails, you get the cause and the query that proves it.

Get started Read the docs

Free when you install it. sigiro sends no data to us.

the loop

Five stages. A repair loop is not complete until all five operate.

Your agent writes the patch. Nothing tells the agent what failed, or if the patch worked. sigiro does two of these five stages.

sigiro

Detect

sigiro compares each service with its own history. You set no thresholds.

your shell

$ sigiro anomalies
SERVICE   SIGNAL       BASELINE   NOW      SHIFT    INCIDENT
checkout  error_rate   0.4%       19.2%    +48σ     inc_4a1c
checkout  latency_p95  180ms      30.0s    +166×    inc_4a1c

2 shifts · 1 incident · since 18:35

sigiro

Diagnose

One call gives ranked results, the suspect deploy, and the query for each result.

your shell

$ sigiro investigate checkout
suspect        deploy 9f2c1ab, landed 18:33
failing        POST /api/orders
failure mode   timeout 246 · server_error 6
errors_total   1284
drill_down_sql SELECT trace_id, duration, status_message
               FROM sigiro_spans WHERE service_name = 'checkout' …

your agent

Fix

Your agent writes the patch. sigiro does not do this step.

your shell

$ 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

your pipeline, on our evidence

Verify

A test that passes does not close the incident. The baseline shows the recovery.

your shell

$ sigiro anomalies --service checkout
no open shifts

error_rate    19.2%  →  0.5%    recovered 19:07
latency_p95   30.0s  →  210ms   recovered 19:07

inc_4a1c closed · 34 min open

your memory

Learn

Keep the fix with the evidence. Use it when the same failure occurs again.

your shell

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

detected   error_rate +48σ, latency_p95 +166×
cause      unbounded upstream call, shipped in 9f2c1ab
fix        5s deadline on the orders client
watch      the same shape on any service calling orders

Detect
Diagnose
Fix
Verify
Learn

sigiroDetect1 of 5

$ sigiro anomalies

SERVICE SIGNAL BASELINE NOW SHIFT INCIDENT

checkout error_rate 0.4% 19.2% +48σ inc_4a1c

checkout latency_p95 180ms 30.0s +166× inc_4a1c

2 shifts · 1 incident · since 18:35

sigiro compares each service with its own history. You set no thresholds.

$ sigiro anomalies
SERVICE   SIGNAL       BASELINE   NOW      SHIFT    INCIDENT
checkout  error_rate   0.4%       19.2%    +48σ     inc_4a1c
checkout  latency_p95  180ms      30.0s    +166×    inc_4a1c

2 shifts · 1 incident · since 18:35

$ sigiro investigate checkout
suspect        deploy 9f2c1ab, landed 18:33
failing        POST /api/orders
failure mode   timeout 246 · server_error 6
errors_total   1284
drill_down_sql SELECT trace_id, duration, status_message
               FROM sigiro_spans WHERE service_name = 'checkout' …

$ 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

$ sigiro anomalies --service checkout
no open shifts

error_rate    19.2%  →  0.5%    recovered 19:07
latency_p95   30.0s  →  210ms   recovered 19:07

inc_4a1c closed · 34 min open

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

detected   error_rate +48σ, latency_p95 +166×
cause      unbounded upstream call, shipped in 9f2c1ab
fix        5s deadline on the orders client
watch      the same shape on any service calling orders

how you use it

Four steps. You connect, ask, correlate, and prove.

You send OpenTelemetry to sigiro. You ask one question. Each step below is one shell session.

Connect

sigiro uses standard OTLP on the standard ports.

your shell

$ docker run -d -p 4317:4317 -p 4318:4318 -p 9999:9999 ghcr.io/sigiroai/sigiro
a3f1c9e2d7b4

$ 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

Ask

sigiro ranks the results and names the deploy that caused the failure.

your shell

$ curl -sX POST localhost:9999/v1/investigate -d '{"service":"checkout"}'
{
  "summary": "two signals moved together on checkout at 18:35",
  "suspect": "deploy 9f2c1ab, landed 18:33",
  "failing": "POST /api/orders",
  "how": { "timeout": 246, "server_error": 6 },
  "errors_total": 1284,
  "query": "SELECT * FROM spans WHERE service = 'checkout' AND ..."
}

Correlate

An agent is a service that calls a model. One trace holds both.

your shell

$ cat models.sql
SELECT model, count(*) AS calls, max(duration) AS slowest
FROM spans JOIN logs USING (trace_id)
WHERE service = 'checkout' AND status = 'error'
GROUP BY model ORDER BY slowest DESC

$ sigiro query --sql-file models.sql
MODEL           CALLS  SLOWEST
gpt-5.2-mini      246    30.0s
claude-haiku-4.6   31     2.1s

Prove

Run the query again. Make the query wider. Show that the result is wrong.

your shell

$ curl -s localhost:9999/v1/investigate -d '{"service":"checkout"}' > answer.json
$ jq .operation_summaries[0] answer.json
{
  "operation": "POST /api/orders",
  "errors": 246,
  "error_rate": 0.19,
  "drill_down_sql": "SELECT trace_id, duration, status_message FROM …"
}

$ jq -r .operation_summaries[0].drill_down_sql answer.json > drill.sql
$ sigiro query --sql-file drill.sql
TRACE_ID       DURATION  STATUS_MESSAGE
9f2c1ab4e77d      30012  context deadline exceeded
3b81fe0a2c56      30004  context deadline exceeded
c40d9e73aa18      29997  context deadline exceeded

Connect
Ask
Correlate
Prove

Connect1 of 4

$ docker run -d -p 4317:4317 -p 4318:4318 -p 9999:9999 ghcr.io/sigiroai/sigiro

a3f1c9e2d7b4

$ 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 uses standard OTLP on the standard ports.

$ docker run -d -p 4317:4317 -p 4318:4318 -p 9999:9999 ghcr.io/sigiroai/sigiro
a3f1c9e2d7b4

$ 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 -sX POST localhost:9999/v1/investigate -d '{"service":"checkout"}'
{
  "summary": "two signals moved together on checkout at 18:35",
  "suspect": "deploy 9f2c1ab, landed 18:33",
  "failing": "POST /api/orders",
  "how": { "timeout": 246, "server_error": 6 },
  "errors_total": 1284,
  "query": "SELECT * FROM spans WHERE service = 'checkout' AND ..."
}

$ cat models.sql
SELECT model, count(*) AS calls, max(duration) AS slowest
FROM spans JOIN logs USING (trace_id)
WHERE service = 'checkout' AND status = 'error'
GROUP BY model ORDER BY slowest DESC

$ sigiro query --sql-file models.sql
MODEL           CALLS  SLOWEST
gpt-5.2-mini      246    30.0s
claude-haiku-4.6   31     2.1s

$ curl -s localhost:9999/v1/investigate -d '{"service":"checkout"}' > answer.json
$ jq .operation_summaries[0] answer.json
{
  "operation": "POST /api/orders",
  "errors": 246,
  "error_rate": 0.19,
  "drill_down_sql": "SELECT trace_id, duration, status_message FROM …"
}

$ jq -r .operation_summaries[0].drill_down_sql answer.json > drill.sql
$ sigiro query --sql-file drill.sql
TRACE_ID       DURATION  STATUS_MESSAGE
9f2c1ab4e77d      30012  context deadline exceeded
3b81fe0a2c56      30004  context deadline exceeded
c40d9e73aa18      29997  context deadline exceeded

Discover. Your agent reads the API description without help.

One public document describes all of the endpoints.

GET /openapi.json · public, no key

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

what you get

No migration. sigiro connects to the tools you use.

Send telemetry from

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

Run it on

Docker
Kubernetes
Fly.io
Railway
Render
Cloudflare

Read by

Claude Code
Cursor
Codex
Anything that speaks SQL

Proof with each result: Each result contains the query that made it.
Nothing to maintain: You set no thresholds and no rules. sigiro uses the history of each service.
Installation in five minutes: One container and two environment variables. You change no code.
Free when you install it: No fees for hosts, seats or data. The hosted service costs approximately $14 each month.
Your data stays with you: You install sigiro on your servers. sigiro sends no data to us.
sigiro monitors sigiro: sigiro sends its telemetry to sigiro. This is how we find our own failures.

faq

The data layer for software that repairs itself.

Five stages. A repair loop is not complete until all five operate.

Four steps. You connect, ask, correlate, and prove.

Discover. Your agent reads the API description without help.

No migration. sigiro connects to the tools you use.

Questions.