Skip to content
TAIP

Products / For AI builders

ImageSphere

Available

An OCI registry with real identity, runtime-mutable policy, and an air-gap story.

ImageSphere is an OCI-native registry that treats access control, identity, and day-2 operations as product features, not afterthoughts. It hosts container images, Helm charts, ML models, and any OCI artifact — with a web UI, group-based admin driven by your IdP, access policies that change at runtime and survive restarts, per-namespace storage quotas with live usage, namespace-scoped robot accounts for CI/CD, immutable tags, and Cosign/Notation verification. Membership is stated in roles people can reason about — viewer, publisher, maintainer — and an explain view answers 'what can this identity actually do here, and why', for a signed-in user, a robot, or an anonymous client. Vulnerability scanning runs on a registry-wide default policy that an admin can exempt per namespace, and the UI is honest about what it knows: it says when scanning is unconfigured rather than showing a reassuring empty list. Helm charts are first-class — detected and rendered with their values.yaml and README, filterable in a charts-only view. One binary, three deployment modes: Kubernetes, air-gapped via the TAIP release flow, bare metal with systemd.

All products

Specification

Version
v1.8.7 — generally available
Protocol
OCI Distribution Spec · built-in /v2/token service
Modes
Kubernetes · air-gapped · bare-metal systemd
Access
Member roles — viewer · publisher · maintainer · plus an explain view for any identity, signed in or not
Scanning
Trivy · registry-wide default severity policy · admin-only per-namespace exemption · database state on the System page
Identity
OIDC SSO · passkeys and a required second factor · API keys and robot accounts with last-used
Governance
Runtime-mutable policies · quotas · immutable tags · audit log
Storage
Local · S3 · GCS
Languages
English · 简体中文

Proof, not promises

See it in one block.

No proprietary SDKs, no rewrites — ImageSphere meets your tools where they already are.

standard clients, governed registry 
$ docker push registry.intra.example/team-a/app:v1
pushed — within team-a quota (412 GiB / 500 GiB)

$ helm push chart.tgz oci://registry.intra.example/team-a
$ cosign verify registry.intra.example/team-a/app:v1
# policies, quotas, and writer groups edited live in the admin UI▌

Pure OCI Distribution Spec — docker, helm, and oras work unmodified. The built-in token service means nothing extra to deploy.

Capabilities

What ImageSphere gives you

01

OCI-native, no proprietary protocols

Standard OCI Distribution Spec — Docker, Helm, ORAS, and any compliant client work unmodified. A built-in /v2/token endpoint means no separate token service to deploy. Helm charts are recognized as charts — values, README, and install hints rendered in the UI, not stored as opaque blobs.

02

Identity from your IdP

OIDC SSO with back-channel logout, passkeys, and a second factor required where a password alone used to be enough. Admin status follows your IdP groups — promoting an admin is a click in your IdP, not a redeploy — and when a local group name collides with one your IdP asserts, the UI says so instead of silently picking. Robot accounts give CI/CD namespace-scoped credentials that outlive any individual, and both they and API keys record when they were last used, so a dormant credential is visible before it is a finding.

03

Access control mutable at runtime

Manage namespace members by role — viewer pulls, publisher pushes, maintainer deletes — and edit groups, quotas, and immutable-tag rules from the admin UI. Changes persist in the metadata store and survive restarts: config file as default, runtime override wins. An explain view answers what a given identity may do in a namespace and which grant said so, anonymous clients included, so 'is this public?' is a question with an answer. An append-only audit log, Prometheus metrics, and an in-app bilingual user guide round out day-2 operations.

04

Vulnerabilities, stated honestly

Trivy scanning behind an admin switch, with a registry-wide default severity policy and an admin-only per-namespace exemption — so an exception is a deliberate, visible act rather than a config file nobody re-reads. Each tag carries a vulnerability panel, the System page inventories every scanner data component and how stale it is, and a version the scanner did not specify is not dressed up as a fix.

05

Air-gap as a first-class deployment

Ships through the TAIP air-gap release flow: chart and images pre-staged, OIDC application registration automated. The pod runs non-root with OIDC and session keys held in Secrets. The same single binary also installs on bare metal with a hardened systemd unit, htpasswd, and TLS.

How it works

From push to pull, governed end to end.

  1. Step 01

    Push any OCI artifact

    Container images, Helm charts, ML models, OPA bundles. Standard /v2/ endpoints; built-in token service for `docker login`.

  2. Step 02

    Identity decides who can do what

    OIDC SSO with passkeys, group-based admin, members managed by role. Ask the explain view what any identity may do here — and off-board in your IdP, and the registry follows.

  3. Step 03

    Pulled, signed, scanned

    Cosign / Notation verification, Trivy scanning under a registry-wide policy with per-namespace exemptions, namespace storage quotas — same plumbing on Kubernetes, air-gap, or bare metal.

Who it's for

Built for these teams

  • Internal platform teams building self-service registries
  • Air-gapped, regulated, and edge deployments
  • Anyone who wants Cosign and CVE scanning without a control plane behind it

Pairs well with

Other builder products

ConsoleX

 Available

Log in, get a governed Kubernetes workspace. No kubectl, no tickets.

On first SSO login every user gets an isolated namespace with quotas, default-deny networking, storage, and a web terminal — provisioned automatically, reconciled continuously. It is also the platform's front door: the token authority every TAIP app trusts, the SSH gateway into every workspace, and the taip CLI the cluster serves itself.

Learn more

DevSpace

 Available

Jupyter or VS Code on a GPU in seconds. Idle environments shut themselves down.

Single-click Jupyter, Marimo, Streamlit, Gradio, and VS Code environments — GPU-ready, isolated per user behind a per-pod auth proxy, shareable with a colleague, and idle-shutdown by default.

Learn more

TrainX

 Available

A template is a shape. Your project fills it. Kubernetes runs the job.

Self-describing templates render straight into UI forms — or your project declares itself in a trainx.yaml and runs on a platform shape. Live quota checks, streaming logs, parsed progress, MLflow-tracked parameters and metrics, resumable runs, and declared outputs you can hand to the next step.

Learn more