Tags

0.17.0-rc.4

0.17.0-rc.4

CLI and monitoring images.

- !422 (#381): mon local-install --instance-jobs / PGAI_INSTANCE_JOBS writes
  COMPOSE_PROFILES into the stack .env, so the outbound collection channel survives
  mon stop/start and mon update instead of being deleted by them. mon health now
  reports a fault when the profile is on and the container is absent, rather than
  '- not enabled'. Disabling a RUNNING container says so explicitly.

- !423 (#382, #348): mon local-install reports the platform's AAS reason instead of a
  bare status. A bare 'HTTP 400' is how platform-all#778 stayed invisible for three
  months across every console-provisioned box.

- !424 (#382): formatPlatformError scrubs by value BEFORE the pattern scrub, so a
  credential echoed with whitespace inside it can no longer have its tail reassembled
  by the flatten.

- !421 (#380): cli:npm:publish is idempotent and waits for the registry, which should
  restore the pgai wrapper alongside postgresai.

Known and shipped: postgresai#383 -- redactTextSecrets stops at the first whitespace,
so a wrapped credential we never sent keeps ~25 chars of its tail. Fix in flight (!425).

0.17.0-rc.3

0.17.0-rc.3

First release carrying the connection inversion: a monitoring instance asks the
platform for work, runs it against its own metric store, and posts the result
back. The platform opens no connection to the instance to collect.

- instance-jobs: the box-side agent (poll, claim, run, submit, sweep), shipped as
  a new postgresai/instance-jobs image. Behind the `instance-jobs` compose
  profile, which nothing enables automatically.
- pgai promql: run a PromQL query on an instance through the platform. Requires
  platform-all's v1.instance_query_enqueue / v1.instance_query_result, which are
  deployed as of api/1.2.51; the channel itself stays off until both
  app.settings.instance_jobs_enabled and app.settings.instance_queries_enabled
  are turned on.
- cli: issues list shows open issues by default and renders status as
  open/closed (#367).

Nothing changes for an existing instance on upgrade: the compose profile is off,
and an instance is only job-backed when the platform flag is on AND it is
actually polling.

Refs #366, #378

0.17.0-rc.2

Release candidate 0.17.0-rc.2

0.17.0-dev.2

0.17.0-dev.2: hidden-issue create/hide via CLI (MR !415, #365)

0.17.0-rc.1

Release candidate 0.17.0-rc.1

0.17.0-dev.1

0.17.0-dev.1

Global-token org selection reaches the wire again for every org-scoped
command (#357, !412): the scope moved off a module-level binding that
bun build was duplicating, onto a globalThis registry symbol shared by
every emitted copy.

Also adds a regression test that runs the BUILT bundle rather than the
TypeScript source, the layer where the defect lived.

0.17.0-dev.0

Dev prerelease 0.17.0-dev.0: global tokens + explicit org selection (multi-org CLI)

0.16.0

Release 0.16.0

0.16.0-rc.5

Release candidate 0.16.0-rc.5

0.16.0-rc.4

0.16.0-rc.3

0.16.0-dev.2

0.16.0-rc.2

0.16.0-dev.1

0.16.0-dev.1 - mon local-install arms hands-off AAS auto-collection (platform-all 338); includes monitor Grafana-user default, datasource readiness retry, ambiguous-datasource short-circuit, reporter URL fix. Dev build for prod auto-AAS rollout.

0.16.0-dev.0

0.16.0-rc.0

0.16.0-rc.1

0.15.0

0.15.0-rc.8

0.15.0-rc.7