Eine selbst gehostete Installation kann sich über die Plattform selbst aktualisieren (
so geht's) oder auf ihrem Server mit
curl -fsSL https://get.equantic.space/install.sh | sudo shDie Versionshinweise erscheinen auf Englisch, so wie jede Version sie mitbringt.
Fehlerbehebungen
- a paused repository's Pipelines tab reads its workflows as paused, closes #645 #645
- eqs reads only its own arguments under BUN_OPTIONS, closes #633 oven-sh/bun#21496 #633
- eqs runs finds a run by repository and number, closes #634 #635 #635 #634
Neu
- eqs services update, deploy --image --wait, and where a service answers, closes #636 #637 #638 #636 #638 #637 #638 #636 #637 #636
Neu
- pause or resume many repositories at once, and list them by visibility
Fehlerbehebungen
- a build's step names stay inside the BuildKit card
Fehlerbehebungen
- the release bundle publishes with a token minted after its build
Fehlerbehebungen
- the web image traces only the docs it reads, not the whole app
Neu
- a CI job's disk is one budget, like a runner's (ADR-038)
Fehlerbehebungen
- an evicted job says why, even when the kubelet's verdict lands after the step
- every registry 401 challenges, so dotnet restore sends its credentials again
Fehlerbehebungen
- package registries fail closed, and nobody publishes into another workspace's package
Neu
- CI storage keeps to a policy, so the disk never fills
Fehlerbehebungen
- a capacity field left empty means automatic, and says so
Fehlerbehebungen
- the idle sweep looks at a machine before deciding about it
Neu
- a runner VM that wakes for a job and switches itself off
Fehlerbehebungen
- the graph actually routes its edges, and the api boots
Fehlerbehebungen
- a workflow edge goes around the cards it used to cross
Neu
- a CI job's memory grows with the job (ADR-035 F1)
- a deployed version says which pipeline built it, and a run says where it went
- every run of the workspace on one page, newest first
- the console draws the docker push it used to transcribe
- the platform ends runs that nothing else will end
Fehlerbehebungen
- a registered machine's key is written where the api can write
- providers get the api's public URL for their webhooks
- the platform can reach its runner VMs
- the platform keeps room to update itself
Neu
- **cli:** eqs runs cancel, dispatch and rerun
- pick many runs and cancel them, in the product's own dialog
Fehlerbehebungen
- buildx builds with a type=gha cache run on the platform
Neu
- a slim release for x64 and arm64, proven before it publishes
- concurrency queue and job-level groups, with GitHub's rules
Neu
- a concurrency group runs one run at a time, and keeps one waiting
Fehlerbehebungen
- a CI slot is a whole pod, so the scheduler can place it
Fehlerbehebungen
- a CI job gets the room the node really has, and capacity fields take a number and a unit
- a due date shows the day that was picked, in São Paulo too
- a settings section on a phone is a screen of its own, not the desktop squeezed in
Neu
- the item's ⋯ on a phone — duplicate, reassign, due date, link, delete
- the run page — a resizable job list, a graph you can dock beside the console, and matrix names that read
Neu
- an item reads as a screen on a phone, not as a desktop modal
Fehlerbehebungen
- every menu escapes its clipping ancestor, through one anchored primitive
Neu
- a Space and its containers are navigable on a phone (ADR-030 M1b)
- the platform update has a door in Settings, on a phone too
Neu
- a job's workspace sits where GitHub puts it, so a step can write beside its checkout
Fehlerbehebungen
- a job's whole log reaches eqs runs logs and the run page, from its first line
- CI on the cluster claims no job before measuring it, waits for room instead of failing, and keeps the image's PATH
Neu
- every link on the site and the platform leads to a real page
Fehlerbehebungen
- a page that redirects no longer flashes an error on a full load (Next.js 16.3.5), closes react/react#36911 vercel/next.js#78396
Fehlerbehebungen
- Help and Documentation open the product docs in the reader's language
- the password reset email and social sign-in land on the platform, not a 404 on the api
Neu
- every screen of the platform speaks both languages
Fehlerbehebungen
- the arg-secret repair read state another boot pass had not written yet
- the CLI still promised the auto-rollout that env apply replaced
Neu
eqs mq — brokers, with the verbs a broker actually has (ADR-031 D5)- a RunnerHost is a first-class row, and the skip stops lying
- Neo4j, where the ADR's read-only claim turns out to be true (ADR-031 D4)
- Redis and MongoDB, and the third answer they forced (ADR-031 D3)
- run jobs on machines the cluster cannot be (ADR-032)
- the platform speaks both languages — shell, settings, repositories and deploy
Fehlerbehebungen
- eqs db never sent its statement — no JSON content type
- the api disconnected a Prisma client it does not own
- the generated Redis password was on the command line
Neu
- run one command in a service, without a terminal (ADR-031 amendment)
Fehlerbehebungen
- the account issuer backfill locked every existing user out
- the mirror handler triggered the duplicate it then worked around
Neu
- MySQL, ClickHouse, and one
tables/describe across the family (ADR-031 D2)
Fehlerbehebungen
- a stateful service that speaks HTTP got no route at all
Fehlerbehebungen
- the marketplace generated secrets that never reached the container
Fehlerbehebungen
- a Service node is a mirror, so nothing else may create one
- run a runner image that only exists on this daemon
Neu
www on any custom domain, per domain and at any time- show when a service has env vars it has not received, and apply them
Fehlerbehebungen
- the two things the first live conformance run found
Fehlerbehebungen
defaults: run: was parsed at both levels and applied at neitherdocker run -v $PWD:/src mounted an empty directory, and said nothinggithub.event_name said push for every run, and dispatch inputs were emptyshell: ran the shell it felt like, not the one the step asked for
Neu
github.event — the delivered webhook payload, which was simply absent- Docker actions run, instead of failing on purpose
Fehlerbehebungen
- redirect HTTP to HTTPS on the ingress, ACME challenge exempt
- the runtime token's
ac claim, and job names that showed their expression
Fehlerbehebungen
- an action we cannot run fails the step instead of finishing green
- export GITHUB_ACTION, without which an App token cannot reach the remote
- take the producer out of the pipe in two
| grep -q conditions
Fehlerbehebungen
- stop the web-deploy step falling back to the API token for the registry
- take the DinD bridge MTU from the pod network, not from an assumption
Fehlerbehebungen
- a service principal can mint a registry token (was a 500)
Fehlerbehebungen
- provisioning a user's first workspace no longer races itself
Neu
- a resizable navigation rail, plus two Docs editor fixes
Fehlerbehebungen
- EQS_RUNNER_DIND never reached the api, so docker-in-job could not be turned on
Fehlerbehebungen
- honour actions/checkout inputs, provide secrets.GITHUB_TOKEN, name the platform
Neu
- Actions cache v2 (twirp CacheService), v1 still fully served
- eqs secret and eqs var — CI secrets and variables from the CLI
Fehlerbehebungen
- an update no longer reports failure while it converges
Neu
- reveal the node a deep link points at in the Spaces tree
Neu
- duplicate and move nodes from the Spaces tree
Neu
- right-click menu on the Spaces tree
Fehlerbehebungen
- **release:** the bundle could not read its own changelog
Fehlerbehebungen
- **platform:** the update modal read its changelog as text, so markdown showed through
- **release:** an issue reference is digits — hex colours are not issues
Leistung
- **platform:** answer the click — skeletons per area, pending nav, one less round-trip, closes #0a0b0e #0f1014
- **repositories:** count workflows with the list, not once per card
Fehlerbehebungen
- **runner:** import IdentityModule, so the capacity routes' guard can resolve
Neu
- **runner:** let the operator tune the CI capacity, and wait instead of starting too small
- **runner:** tell a reader when a step is starved, not hung
Fehlerbehebungen
- **runner:** let a CI step signal its own children — AppArmor was denying it
Fehlerbehebungen
- **catalog:** one catalog, not two — the admin's copy shipped services that could not deploy
- **deploy:** let a marketplace service say how it starts, so MinIO can run at all
Neu
- **platform:** notifications that carry the action, and one primitive to raise them
Neu
- **platform:** activity bell + reactive Space tree for deploy actions
Neu
- **runner:** derive the CI budget from real cluster availability, live
- **runner:** say when a step was killed by the job memory limit instead of a bare exit 137
Fehlerbehebungen
- **runner:** survive an installation token that expires mid-job (401 on action fetch)
Neu
- **runs:** stopwatches, tail-following logs and a run view that follows the pipeline
Fehlerbehebungen
- **runner:** rebuild the full contexts when reconstructing a reusable sub-job's scope
Fehlerbehebungen
- **release:** read EQS_LICENSE_PUBLIC_KEY from .env when packaging the bundle
Fehlerbehebungen
- **runner:** preserve the container env through sudo for non-root steps (kubernetes substrate)
Fehlerbehebungen
- **runner:** env expression context sees declared workflow/job env, not only GITHUB_ENV carries
Fehlerbehebungen
- **setup:** write EQS_DISPATCHER into the in-cluster api env — dispatcher was silently off
Neu
- **deploy:** deploy the TCP gateway on install and on update (ADR-028 G2)
- **deploy:** TCP SNI gateway — one port for every database (ADR-028 G1)
- **runner:** Kubernetes pod-per-job execution substrate — MVP (ADR-029 K2)
- **runner:** per-job limits + cluster-derived CI capacity (ADR-029 K4)
- **runner:** private per-job DinD sidecar for in-job docker (ADR-029 K5)
- **runner:** services, container jobs and registry auth on the pod substrate (ADR-029 K3)
- **setup:** pods are the default CI substrate — dind retired on install and update (ADR-029 K6)
Neu
- **deploy:** give a public TCP service its own hostname (ADR-028 G0)
Fehlerbehebungen
- **admin:** read the CR token from .env, like every sibling script does
- **deploy:** show where a public TCP service answers, and stop cutting its neighbours
Fehlerbehebungen
- **deploy:** reject an invalid source CIDR at the save, not mid-rollout
Neu
- **license:** revalidate the licence from the UI, and keep the key across every update path
Fehlerbehebungen
- **setup:** never finish an install without an admin, and honour offline serials at runtime
Fehlerbehebungen
- **setup:** accept a signed serial's length at the API boundary
- **setup:** actually bake the license public key into the bundle
Neu
- **setup:** offline, signature-verified license serials
Neu
- **deploy:** restrict a public TCP port to chosen origins (F2)
Neu
- **deploy:** expose TCP services on a dedicated public port
Fehlerbehebungen
- **deploy:** stop offering a public domain to TCP services
Neu
- **work:** startDate on Item, so a Gantt bar has two ends
Neu
- **catalog:** curate connectability in the admin + API-key services
Fehlerbehebungen
- **catalog:** backfill connection metadata on catalogs seeded earlier
Neu
- **deploy:** Connect dialog with the service's real connection string
Fehlerbehebungen
- **deploy:** pull vcluster's rewrite-hosts init container from a registry that answers
Fehlerbehebungen
- **update:** keep EQS_VERSION in lockstep with the image on self-update