sig is an independent CLI for SigNoz. Find the logs behind an error, inspect a slow trace, or query a metric without leaving your workflow. It connects directly to your SigNoz API and returns JSON that works equally well in a shell pipeline or an agent's tool call.

• Investigate across signals. Search logs and spans, fetch trace waterfalls, calculate statistics, and run PromQL.
• Discover as you go. Find field names, observed values, and available metrics before building a query.
• Bring your agent. Machine-readable commands and offline query recipes are built in. No MCP server is needed.

Install

CLI

go install github.com/sprisa/sig@latest # with Golang (no auto update)
npm install -g @sprisa/sig # with npm (no auto update)

Check either installation:

sig version

Installation help · Build from source

Agent skill

Install the skill for your AI agent with skills (requires Node.js and npx):

npx skills add sprisa/sig

The skill includes CLI installation instructions, so you can start here even if you haven't installed sig yet. It also teaches the investigation workflow and when to load the bundled query recipes.

Quick start

You'll need your SigNoz URL and a service-account API key. Replace the example URL with your instance:

sig auth login --url https://signoz.example.com
sig auth status
sig logs search --since 15m

Login prompts for the key and saves it in your OS keychain. Your first login creates the default context, so subsequent commands already know where to connect.

Using CI, a headless machine, or an existing MCP configuration? You can supply SIGNOZ_URL, SIGNOZ_API_KEY, and optional SIGNOZ_CUSTOM_HEADERS through the environment instead. See authentication and configuration.

Everyday queries

Find the logs you need

sig logs search --where "body CONTAINS 'timeout'" --since 1h --limit 20

Not sure what to filter on? Discover the fields and values in your instance:

sig logs fields --search service --since 1h
sig logs values severity_text --field-context log --since 1h

Log query recipes →

Get the big picture

Count matching records or group them into a trend instead of downloading every row:

sig logs aggregate --aggregation 'count()' --since 1h
sig logs aggregate --aggregation 'count()' --group-by severity_text --step 1m --since 1h

Aggregation recipes →

Follow a trace

Search returns spans. Use a trace ID from the results to retrieve its waterfall:

sig traces search --where "has_error = true" --since 1h --limit 20
sig traces get "$TRACE_ID"

Trace query recipes →

Query your metrics

Discover a metric name, then query it with PromQL. Replace the example expression with one that matches your instrumentation:

sig metrics list --search http --since 1h
sig metrics query 'sum(rate(http_requests_total[5m]))' --since 1h --step 1m

Metric query recipes →

Use explicit bounds for an incident window (replace these example times):

sig logs search --start 2026-01-01T12:00:00Z --end 2026-01-01T13:00:00Z

Fetch a few pages or resume with meta.next_page_token from a previous response:

sig logs search --since 1h --limit 100 --pages 3
sig logs search --page-token "$PAGE_TOKEN"

For a native SigNoz v5 query you've saved to a file:

sig query preview --file query.json
sig query run --file query.json

Query behavior, limits, and native examples →

JSON that fits your workflow

Successful commands write {data, meta?} to stdout. Errors write {error} to stderr and return a nonzero exit code. Help is plain text.

For example, use jq to inspect the metadata for a search:

sig logs search --since 1h --limit 20 | jq '.meta'

Check warnings and completeness before treating a result as exhaustive. Output shapes, pagination, and exit codes explain what each command returns.

AI agents

Use the skill install command above, or add skills/sig/SKILL.md through your agent's usual skill mechanism. The skill points to focused recipes instead of loading a full manual on every request.

Agents can also discover the installed interface directly:

sig agent schema logs aggregate
sig agent recipes
sig agent recipes workflow

Schema and recipes work offline. Everyone gets the same JSON interface; no agent-mode flag or environment detection is required.

Learn more

| Looking for… | Start here | | --- | --- | | Headless auth, multiple contexts, or proxy headers | Configuration | | Time windows, pagination, output contracts, or native JSON queries | Querying | | Field contexts, units, and investigation techniques | Query recipes | | Credential handling and the trust model | Security | | Building, testing, architecture, or contributing | Contributing |

Run sig --help or sig <command> --help for command details.

Compatibility: sig targets SigNoz v0.132.0's APIs. Compatibility with other versions needs verification. See compatibility notes.

Licensed under MPL 2.0.