dgem v0.4.1

Install the CLI

Download a prebuilt static archive (dgem, embedded Decision Studio, templates/, LICENSE, and README.md) and verify its SHA-256 checksum:

OS="$(uname -s | tr '[:upper:]' '[:lower:]')"
ARCH="$(uname -m | sed 's/x86_64/amd64/;s/aarch64/arm64/')"
curl -fsSLO "https://github.com/GoogleCloudPlatform/dgem/releases/download/v0.4.1/dgem_v0.4.1_${OS}_${ARCH}.tar.gz"
curl -fsSLO "https://github.com/GoogleCloudPlatform/dgem/releases/download/v0.4.1/SHA256SUMS"
tar -xzf "dgem_v0.4.1_${OS}_${ARCH}.tar.gz"
./dgem --version

Or install with Go 1.24+:

go install github.com/GoogleCloudPlatform/dgem@v0.4.1
dgem --version

Serving image. No prebuilt images are published yet. Build the serving image from this tag in your own Google
Cloud project with GCP_PROJECT=<PROJECT> make cloudrun-build, then deploy it to Cloud Run GPU or Vertex AI (see
Build the Serving Image and the deploy guides).

---

• Gateway /v1/systemone latency-tail instrumentation and connection pooling (cmd/serve.go, cmd/root.go).
  v1ProxyHandler (POST /v1/systemone and /v1/chat/completions) now emits sequential child spans under
  dgem.gateway.proxy (dgem.gateway.read_body, dgem.gateway.resolve_backend, dgem.gateway.prepare_body,
  dgem.gateway.auth_token, dgem.gpu.proxy_forward, and dgem.gateway.write_response) with request/response byte
  counts, Vertex status probe telemetry (dgem.vertex_status.cache_hit, dgem.vertex_status.source), token cache
  hit/miss (dgem.token.cache_hit), and httptrace upstream connection reuse (dgem.conn.reused, dgem.conn.was_idle,
  dgem.conn.idle_ms). Upstream proxy calls now share a pooled http.Transport (MaxIdleConns: 128,
  MaxIdleConnsPerHost: 64, IdleConnTimeout: 90s) instead of http.DefaultTransport (MaxIdleConnsPerHost: 2).
• Stale-while-revalidate Vertex endpoint status and :rawPredict fast health probe (cmd/serve.go).
  inspectVertexEndpointState serves cached deployed status immediately within a 90s grace window after its 30s
  fresh TTL while refreshing asynchronously in the background (stale_cache), deduplicates concurrent refreshes per
  endpoint ID, extracts project/region directly from Vertex endpoint URLs before querying metadata, and probes fast
  data-plane {"health": true} on :rawPredict endpoints (sending X-DGem-Key when DGEM_SERVER_KEY is configured)
  before falling back to the Vertex endpoints.get control-plane API.
• Prebuilt CLI release archives and go install version reporting (.github/workflows/release.yml, cmd/version.go).
  Pushing a v*.*.* release tag now builds Decision Studio (studio/dist), cross-compiles static CGO_ENABLED=0 dgem
  binaries for linux/amd64, linux/arm64, darwin/amd64, and darwin/arm64 stamped with the release tag and commit
  revision, packages dgem, templates/, LICENSE, and README.md with SHA256SUMS, and attaches them to the
  GitHub Release (creating a draft release if needed). Module-mode go install github.com/GoogleCloudPlatform/dgem@vX.Y.Z
  builds also fall back to Go's embedded module BuildInfo version instead of reporting "dev".
• Interactive documentation landing page and .github/CODEOWNERS (docs-site/src/pages/index.astro, docs-site/public/landing/).
  Adds an interactive GM3 documentation home page featuring a live slot-denoising simulator, Shannon hesitation lab,
  and copy-ready policy workbench, plus repository .github/CODEOWNERS.