Skip to content

v0.1.0

Latest

Choose a tag to compare

@github-actions github-actions released this 24 Aug 07:56
· 7 commits to main since this release
99a464c

Streamweld v0.1.0

The first public release of Streamweld, an OpenAI-compatible durability layer
for streaming responses from self-hosted LLM inference servers.

Included

  • A Go proxy with memory and Redis journals, commit-before-visibility SSE
    delivery, exact Last-Event-ID replay, idempotent stream creation, bounded
    readers, and explicit stop semantics.
  • A Kubernetes operator and Helm chart for InferenceRoute and
    DurabilityPolicy resources, compatibility-aware routing, and backend drain
    coordination.
  • Guarded producer continuation with model, tokenizer, template, request-shape,
    token-budget, structured-output, tool-call, terminal-state, and seam checks.
    Unsafe continuation is refused.
  • streamweldctl commands for compatibility checks, stream inspection, drain
    operations, and deterministic benchmark verification.
  • @streamweld/client for browser, Node.js, Deno, Bun, and edge runtimes, plus
    @streamweld/ai-sdk for Vercel AI SDK v5.
  • Prometheus metrics, OpenTelemetry traces, structured logs, an optional
    ServiceMonitor, and a Grafana dashboard.
  • Multi-architecture proxy and operator images for Linux AMD64 and ARM64.

Install

Install the Kubernetes components:

helm upgrade --install streamweld oci://ghcr.io/satwiksps/charts/streamweld \
  --namespace streamweld-system \
  --create-namespace \
  --version 0.1.0 \
  --wait --timeout 3m

Install the TypeScript client:

npm install @streamweld/client

Vercel AI SDK v5 applications can add the transport adapter:

npm install @streamweld/ai-sdk ai@^5

Prebuilt streamweld-proxy, streamweld-operator, and streamweldctl archives
for Linux, macOS, and Windows are attached below. SHA-256 checksums are provided
in checksums.txt.

Container images are available as
ghcr.io/satwiksps/streamweld-proxy:0.1.0 and
ghcr.io/satwiksps/streamweld-operator:0.1.0.

Compatibility

  • Kubernetes 1.27 or newer
  • Helm 3.14 or newer
  • Node.js 20 or newer for the TypeScript packages
  • OpenAI-compatible streaming inference backends

The default memory journal supports one proxy replica. Configure Redis before
scaling the proxy horizontally. Run streamweldctl doctor against the exact
backend image, model, and tokenizer before enabling producer migration.

Streamweld is a self-hosted durability layer, not a model server. Version 0.x
APIs and v1alpha1 Kubernetes resources may change.

Documentation ·
Security policy