SparkRoute v0.0.2

SparkRoute v0.0.2 adds opt-in automatic recovery for unhealthy SparkRun-managed workloads.

• Restart a persistently failing owned workload, or stop it until demand returns.
• Require qualifying failed half-open probes and a minimum unhealthy duration before recovery begins.
• Drain shared request leases, use bounded retries and backoff, and stop automatic attempts when the restart budget is exhausted.
• Configure recovery in the recipe deployment editor and inspect recovery status or retry manually from the deployment overview.

Recovery is disabled by default. See the workload recovery guide for configuration, defaults, and behavior with idle shutdown.

Release artifacts include binaries for Linux, macOS, and Windows on amd64 and arm64, corresponding source, license notices, and SHA-256 checksums. The container is published as ghcr.io/sparksq/sparkroute:0.0.2.

Full changelog

SparkRoute v0.0.1

SparkRoute v0.0.1 is the first public release of the standalone AI gateway and its embedded administration console.

• Route OpenAI-compatible, Anthropic, Gemini, and Bedrock requests through virtual models, aliases, and configurable routing policies.
• Manage providers, deployments, routing, credentials, and named operator configuration presets in the console.
• Discover and activate local models through the optional Sparkrun integration, with workload lifecycle controls in Overview.
• Configure guardrails, optional PII handling, usage accounting, and saved traces.

Download the archive for your operating system and architecture below. Linux, macOS, and Windows are supported on amd64 and arm64. Verify the archive against checksums.txt; each archive includes the executable, exact source identity, and license notices. The corresponding source archive is included.

The container is ghcr.io/sparksq/sparkroute:0.0.1 for Linux amd64 and arm64. Sparkrun's subprocess integration requires a host installation; the routing container does not include Sparkrun.

SparkRoute is distributed under AGPL-3.0-only, with upstream and dependency notices retained. See release documentation for build and verification details, and the on-demand model documentation for setup and the known limitation with sub-second idle intervals.