Grafana GCX (CLI)

grafana/gcx last check 35 releases
Notes
no notes yet
Release notes
view on github
1.7.1

v1.7.0 was tagged but never published: its image build failed. v1.7.1 is the first published release with the v1.7.0 changes, listed below.

  • Build release binaries with Go 1.26.10 and update golang.org/x/net to v0.60.0, picking up standard library security fixes and five HTTP/2 fixes (GO-2026-6603, -6610, -6611, -6612, -6617) (#1625)
  • Fix the container image build, which stopped the v1.7.0 release before it published binaries or images (#1623)

Changes from v1.7.0

Breaking changes

  • Some failures now exit with different codes. HTTP 401/403 from product commands, Assistant helpers and gcx api, and missing Cloud or Synthetic Monitoring credentials exit 3 instead of 1. Unknown flags and missing or extra positional arguments exit 2 instead of 1. Error summaries use a fixed vocabulary (Authentication failed, Authorization failed, Resource not found, Resource conflict, Invalid command usage). Update scripts that branch on exit 1 or match the old error text. (#1483)
  • Contact points, notification policies, templates and mute timings that gcx writes now stay editable in the Grafana UI by default. Pass --allow-ui-updates=false to keep them provisioned. Updating a contact point that is already provisioned requires --allow-ui-updates=false; without it Grafana rejects the update. Alert rules are unchanged. (#1553)
  • gcx synthetic-monitoring checks status and checks get --show-status now read from the Synthetic Monitoring datasource and require Synthetic Monitoring app v1.64.0 or later. SUCCESS (JSON success) is now reachability over 3 hours instead of a 5-minute probe_success average, so some checks may change between OK and FAILING. --datasource-uid is hidden and ignored; it is still accepted, with a warning. (#1489)
  • gcx logs query and gcx datasources loki query (including generic datasources query against Loki) now reject metric queries such as rate(...) in every output format, instead of printing metric values as log lines. Run metric queries with gcx logs metrics. (#1580)
  • gcx kg notifications upsert now rejects unknown or misspelled fields, and input containing more than one YAML document. Both previously succeeded, dropping the unknown key or the extra documents. It now also accepts the output of notifications get and list directly. Correct field names in saved notification files before you upgrade. (#1571)

New commands and improvements

  • Add Assistant Watchers: inspect, export, create, update and lifecycle commands
  • Watchers: run history, version restore, calibration and notification routing
  • Add alerting time interval and template group workflows
  • Add kg graph queries and threshold commands; deprecate legacy entity queries
  • Improve kg graph query and meta schema help, with --jq examples
  • Add synthetic-monitoring suggestions list
  • Add CSV output (-o csv) for Prometheus, Loki, SQL, tabular, Cloud, search-metrics, Tempo and Pyroscope queries
  • Add datasources loki query-patterns and logs query-patterns
  • Render -o graph for Loki log queries as a log-volume histogram
  • Describe the profiling workflow in profiles help
  • Select the active context with the GCX_CONTEXT environment variable
  • Detect Hermes Agent and OpenClaw to turn on agent mode
  • Publish a container image to ghcr.io and Docker Hub, signed with keyless cosign
  • Make SLO, Synthetic, Alerting, k6, Fleet, stack and telemetry tables consistent
  • Fix decoding of SLOs that use grafanaQueries
  • Fix irm oncall schedules list-final-shifts ignoring --start
  • Fix OAuth login callback by using an OS-assigned port
  • Fix config validation errors and legacy mTLS with env-only auth or tokens
  • Fix Linux keychain prompting for unlock in headless sessions
  • Fix instrumentation sending a Pyroscope URL ending in /ingest
  • Sandbox recovers from a broken compilation cache and reports why
  • Docs: CLI reference now ships with releases; added Grafana MCP comparison