Grid v0.1.3

Grid v0.1.3 changes

• Added separate health and TLS endpoints for Kubernetes health checks.
• Added selectable provider scoring strategies: noMetrics, queueDepth, and kvCachePressure.
• The noMetrics default disables dynamic metric preference while retaining health and routing policies.
• Improved VCR pressure-routing demo and metrics handling.
• Added registry-backed metrics demo support.
• Added maintainer-triggered PR E2E workflows with quick/full modes.
• Improved llm-d image loading and optional metrics transport.
• Published official Grid v0.1.3 operator, overlay-sync, mock-provider, AI rollup images, and Helm charts.

Container Images

ghcr.io/praxis-proxy/grid-ai-rollup:v0.1.3@sha256:e5bdea9071f76533ab61714eeef6067c7449d0e9d811c221aa940cf3cdf9719f
ghcr.io/praxis-proxy/grid-operator@sha256:c213128e6f70f528e7238b695f2993ffe4224eac38dc893161423e3feee53f35
ghcr.io/praxis-proxy/grid-mock-providers@sha256:e0ab1e839f07af0c6bb170a7653079cf7c96f91f10ceadabd9af669ce72f1fef
ghcr.io/praxis-proxy/grid-overlay-sync@sha256:b90169e98593597b445e26eacaf5a08a79254c91134f4b9161002da5331f463c

Helm Chart

oci://ghcr.io/praxis-proxy/charts/grid-operator --version 0.1.3
oci://ghcr.io/praxis-proxy/charts/grid-operator@sha256:6eb4575b0a15be75c6fc26e3db23a76f3a8a685b6cb54350a853bf72ce3b79b1
oci://ghcr.io/praxis-proxy/charts/praxis-gateway --version 0.1.3
oci://ghcr.io/praxis-proxy/charts/grid-site --version 0.1.3
oci://ghcr.io/praxis-proxy/charts/grid-mock-providers --version 0.1.3

Install:

helm install grid-operator \
  oci://ghcr.io/praxis-proxy/charts/grid-operator \
  --version 0.1.3 \
  --namespace grid-system --create-namespace

The Grid images were built from 068c68958456ad8bcba08d651213a3d8566a4c30. The Praxis AI
rollup was built from
praxis-proxy/ai@c68149502d991d51daaff8d6bf96c18704aeb38d and
praxis-proxy/praxis@0d3347c634ab4e09d882c02bf01767a1667c82db.
Every reference above is immutable.

What's Changed

• feat: separate endpoint and tls support for healthchecks by @mvazquezc in #29

New Contributors

• @mvazquezc made their first contribution in #29

Full Changelog: v0.1.2...v0.1.3

Grid v0.1.2

Container Images

ghcr.io/praxis-proxy/grid-ai-rollup@sha256:95132eb39c0f568b5361a250002979c5063db427ff0fb63b59a93146fcb7ad31
ghcr.io/praxis-proxy/grid-operator@sha256:578e26129d1b281e664b5261b345b76cd38844a58a400adc48f3d7d9c58518e5
ghcr.io/praxis-proxy/grid-mock-providers@sha256:70315c3aa58747b92d5e3d6c8406d13606a93c486c78d0cd6eaf655be2ede7bf
ghcr.io/praxis-proxy/grid-overlay-sync@sha256:2e42bce30673c2363157495ff72c12fe5669159382a102de0d0df7459939018c

Helm Chart

oci://ghcr.io/praxis-proxy/charts/grid-operator --version 0.1.2
oci://ghcr.io/praxis-proxy/charts/grid-operator@sha256:d2c3cea00a609c88872dc17fb3de2d89973399ffa45c6d7b7aebeb2e5046685e
oci://ghcr.io/praxis-proxy/charts/praxis-gateway --version 0.1.2
oci://ghcr.io/praxis-proxy/charts/grid-site --version 0.1.2
oci://ghcr.io/praxis-proxy/charts/grid-mock-providers --version 0.1.2

Install:

helm install grid-operator \
  oci://ghcr.io/praxis-proxy/charts/grid-operator \
  --version 0.1.2 \
  --namespace grid-system --create-namespace

The Grid images were built from f17b2fc6ee9433fd23096164d6c308cbded3b995. The Praxis AI
rollup was built from
praxis-proxy/ai@7341b2d593f9aaafee6b8c3e1be75a2c6ea4108e and
praxis-proxy/praxis@e6001bad6fc42fde26cecca4682f04ddedc57abf.
Every reference above is immutable.

What's Changed

• feat(demo): add MaaS IPP Kind lab by @aslakknutsen in #22

New Contributors

• @aslakknutsen made their first contribution in #22

Full Changelog: v0.1.1...v0.1.2

Grid v0.1.1

Highlights

• Adds the production Helm workflow for adding, removing, and restoring inference providers.
• Documents in-cluster, existing-Service, and external HTTPS provider integration.
• Adds provider workflow, multi-provider, NetworkPolicy, values-precedence, and removal validation.
• Completes the runnable three-site combined-site demonstration.
• Proves local preference, remote fallback, session behavior, provider add/remove/re-add, deterministic identity, overlay convergence, security boundaries, operator recovery, and cleanup.
• Pins the Getting Started demo to the compatible project-owned image digests.

Temporary Praxis AI and Praxis Core Forks

Grid v0.1.1 temporarily builds the grid-ai-rollup image from development revisions of Praxis AI and Praxis Core. The required Grid integration changes have not all reached compatible upstream releases, so building from released upstream versions would omit functionality exercised by the Grid demos and Helm deployments.

The rollup remains a single gateway image. It does not require separate Praxis AI and Praxis Core containers. The forks are source dependencies used to produce that combined image until the changes are merged and released upstream.

Required Praxis AI Features

• Grid-aware intelligent provider selection from operator-generated routing overlays.
• Versioned overlay validation and hot reload with last-known-good retention.
• Provider-hop routing context and serving-revision attribution.
• Provider-boundary authorization for exact candidate, model, path, and cluster mappings.
• Final-hop provider credential injection and caller authorization replacement.
• Session affinity across local selection, remote fallback, and restoration.
• Optional external HTTPS provider routing, including OpenAI.

Required Praxis Core Features

• Per-cluster HTTP authority override for external HTTPS providers.
• Independent TLS SNI and HTTP authority configuration.
• mTLS client-certificate enforcement.
• Peer identity and trust-policy enforcement.
• Load-balancer behavior required by consumer-to-provider and provider-to-backend routing.
• Runtime and configuration support used by the Praxis AI Grid filters.

The temporary source revisions are:

praxis-proxy/ai@7341b2d
praxis-proxy/praxis@e6001ba

These forks should be retired once equivalent Praxis AI and Praxis Core changes are merged, released, and validated together. The immutable grid-ai-rollup digest records the exact compatible implementation used by Grid v0.1.1.

Runnable Grid Demonstrations

Grid now includes automated, evidence-backed demonstrations for its primary deployment topologies. See the Grid demos page (https://github.com/praxis-proxy/grid/tree/main/demos) for prerequisites, topology diagrams, commands, and validation details.

• Global Ingress: Demonstrates external traffic entering through healthy edge gateways before Grid selects an eligible provider.

• Workload Inference: Demonstrates cluster-local workloads using consumer gateways with dedicated provider clusters, local preference, and remote fallback.

• Combined Sites: Demonstrates three clusters that each run separate consumer and provider gateways, including local preference, cross-site fallback, session behavior, provider add/remove/ re-add, overlay convergence, security enforcement, recovery, and cleanup.

• Optional external providers: Supports opt-in live OpenAI routing while keeping credentials isolated to the selected provider gateway.

• Automated runtime proof: Demo assertions validate mTLS, peer authorization, credential replacement, backend isolation, routing attribution, provider lifecycle, stable identity, operator recovery, and sanitized evidence.

Container Images

ghcr.io/praxis-proxy/grid-ai-rollup@sha256:95132eb39c0f568b5361a250002979c5063db427ff0fb63b59a93146fcb7ad31
ghcr.io/praxis-proxy/grid-operator@sha256:654d9079e13c80e7891dcdd2eed52901ebd733833ae02d776a69a4170c00d9bb
ghcr.io/praxis-proxy/grid-mock-providers@sha256:60c9ac29782b2ce6c99eb4d82494bd10280ef06b453752486f6933927547d333

Helm Charts

oci://ghcr.io/praxis-proxy/charts/grid-operator --version 0.1.1
oci://ghcr.io/praxis-proxy/charts/grid-operator@sha256:2c5ecf8d23846aeb403616e6501b72be872a8f59006536af9f87cad7a409ed7e
oci://ghcr.io/praxis-proxy/charts/praxis-gateway --version 0.1.1
oci://ghcr.io/praxis-proxy/charts/grid-site --version 0.1.1
oci://ghcr.io/praxis-proxy/charts/grid-mock-providers --version 0.1.1

helm install grid-operator \
  oci://ghcr.io/praxis-proxy/charts/grid-operator \
  --version 0.1.1 \
  --namespace grid-system --create-namespace

Full Changelog: v0.1.0...v0.1.1

Grid v0.1.0

Grid v0.1.0

Grid v0.1.0 provides the first reproducible implementation of distributed, multi-cluster inference routing for Praxis.

Functionality

• Kubernetes operators and CRDs for defining Grid networks and sites.
• Decentralized site discovery and partition-tolerant state propagation.
• Active/active Praxis edge gateways behind a stable ingress endpoint.
• Multiple independently selectable providers within a site.
• Health, capacity, locality, freshness, and admission-aware routing.
• Session affinity with provider drain behavior.
• Provider and edge failure detection, withdrawal, and recovery.
• mTLS site trust, provider authorization, and credential isolation.
• Versioned routing overlays with observable revision propagation.
• Hot configuration reload without restarting the Praxis edge.
• Last-known-good routing when an invalid overlay is received.
• Machine-readable evidence for routing, security, restart, and soak tests.

Demo

The Grid global ingress demo (https://github.com/praxis-proxy/grid/tree/v0.1.0/demos/grid-glb-demo) provides a reproducible local environment using Kubernetes-in-Docker clusters.

The demo validates the following functionality through live runtime assertions:

• Site discovery
• Routing-overlay propagation
• Active/active edge routing
• Multiple providers per site
• Secure provider boundaries
• Session affinity
• Provider drain
• Edge withdrawal and recovery
• Grid restart recovery
• Invalid-overlay protection
• Sustained traffic validation

Praxis AI Dependency

The demo currently uses a temporary Grid-specific fork and rollup of Praxis AI. It combines functionality from open Praxis AI changes that have not yet landed in a standard Praxis AI release:

• Intelligent Grid routing (praxis-proxy/ai#339)
• Routing-overlay hot reload and observable envelope (praxis-proxy/ai#540)
• Secure provider-routing boundary (praxis-proxy/ai#386)

The rollup is published separately as ghcr.io/praxis-proxy/grid-ai-rollup and pinned by immutable digest in the demo documentation. It is intended for integration and demonstration until the underlying changes are reviewed, merged, and included in an upstream Praxis AI release.

Known Limitations

• The included topology models multiple clusters within one region.
• Global ingress uses a local GTM emulator; production DNS or Route 53 integration is not included.
• Mock inference providers are for demonstration and validation only.
• Cross-region routing, regional affinity and locking, and direct public-cloud provider routing remain outside this release.