🚀 Sparge v1.0 is open source — star us on GitHub and join the waitlist

Changelog

What's new

v1.0.0

April 27, 2026

major

First Public Release

Sparge v1.0 is now open source. Model-agnostic, observable by design, deployable in minutes.

  • ◈Nine reliability engines (Apache 2.0): SLO Engine, Context Scorer, Lineage Graph, Drift Detector, Cost Router, AI Narrator, Fleet Registry, Red Team Engine, Audit Log
  • ◈Multi-model AI via LiteLLM — 100+ providers including local Ollama (no API key required)
  • ◈Full self-observability: 18 Prometheus metrics, OTEL traces to Grafana Tempo, structlog JSON to Loki
  • ◈One-liner installer: curl -fsSL https://get.sparge.ai | sh
  • ◈Kubernetes operator with sidecar injection (failurePolicy: Ignore)
  • ◈AgentSLO + AgentFleet Kubernetes CRDs
  • ◈Docker Compose: 8-service observability stack (OTEL + Prometheus + Grafana + Tempo + Loki + Alertmanager)
  • ◈Pre-built Alertmanager rules for SLO breach, context quality, and LLM cost alerts
  • ◈Enterprise integrations: ServiceNow (FastMCP 3.2), Jira, PagerDuty, Slack
  • ◈52 tests across 6 test modules

v0.9.0

March 15, 2026

minor

Multi-Model AI Provider

Replaced vendor-specific AI integration with LiteLLM-based abstraction layer.

  • ◈LiteLLM integration supporting 100+ LLM providers
  • ◈Per-use-case model selection: SPARGE_NARRATOR_MODEL, SPARGE_JUDGE_MODEL
  • ◈Automatic fallback chains on provider failure
  • ◈Ollama local model support — fully air-gapped operation
  • ◈Cost tracking per provider per model

v0.8.0

February 20, 2026

minor

Self-Observability Layer

Sparge now instruments itself with the same rigour it applies to monitored agents.

  • ◈18 Prometheus metrics at /metrics endpoint
  • ◈OpenTelemetry auto-instrumentation (FastAPI + engines)
  • ◈structlog JSON structured logging with correlation IDs
  • ◈@timed_engine decorator for engine performance tracking
  • ◈Pre-built Grafana dashboards (Platform Overview, AI Cost Intelligence)
  • ◈Alertmanager rules pre-configured

v0.7.0

January 30, 2026

minor

Context Quality Scorer

Three-dimension context quality scoring detects root cause of >50% of production failures.

  • ◈Freshness (45%) + Completeness (35%) + Consistency (20%) scoring
  • ◈12 enterprise data type refresh expectations
  • ◈Schema drift detection between observations
  • ◈Under 2ms scoring latency — no external API calls
  • ◈Recommendations generated on quality degradation

Ready when you are

Your agents are deployed.
Are they reliable?

Find out in 5 minutes with the open source core. Any LLM provider. Fully observable from day one.

Open Source on GitHubGet Early Access

Model-agnostic. Observable by design. Deployable in minutes.