- Hardened the public release packaging and runtime version metadata.
- Corrected Docker Compose Redis configuration to use
SENTINELFLOW_REDIS_URL. - Release builds now inject version metadata into
internal/version.
- Release artifacts are built and attested by the GitHub Actions release pipeline.
- Intelligent AI core with intent, complexity and capability analysis.
- Adaptive routing and outcome learning primitives.
- Agent runtime with approvals, cancellation and tool execution.
- Tenant/project-isolated memory and knowledge/RAG foundations.
- AI Security 2.0 with risk-based ALLOW/REVIEW/BLOCK decisions.
- Predictive FinOps, semantic AI observability and evaluation primitives.
- Prompt versioning, multimodal request parts and AI governance.
- Plugin registry, global routing, self-healing, Zero Trust AI and autoscaling foundations.
- Dedicated V5 CI, regression harness and release documentation.
- V5.0.1 is the current published GA maintenance release.
- Production deployment remains environment-specific; follow the V5 production-readiness checklist before operating at scale.
- V4 foundation, distributed runtime, policy lifecycle, routing, FinOps, observability and enterprise contracts.
- Developer portal/playground and Go/Python SDKs.
- Docker Compose, Helm, Terraform, SBOM/security CI and production-readiness documentation.
- V4 uses explicit PostgreSQL migrations and removes the legacy rules.yaml/SQLite configuration path.
- Repository documentation and community health files were refreshed for open-source contributors.
Reescritura completa con arquitectura de 3 planos.
internal/version: single source of truthinternal/config/v3: static/dynamic/secrets separationinternal/storage/postgres: pgx pool + migraciones embebidas- Nueva estructura
cmd/: gateway, controlplane, worker, cli
- Repositorios persistentes: organizations, projects, users, api_keys
- API keys SHA-256 + scopes + TTL + rotation
- RequestNormalizer: OpenAI/Anthropic/Gemini
- AuthMiddleware contra PostgreSQL
- IdempotencyMiddleware
- Streaming SSE real
- Interface Provider con Chat/Stream/Health/Models
- Registry thread-safe
- Health monitor cada 30s
- Circuit breaker por provider
- Adapters: OpenAI, Anthropic, Ollama
- Model Registry con 9 modelos + pricing real
- Scoring: latency 0.35 + cost 0.25 + health 0.25 + quality 0.15
- Filters + Experiments (A/B, canary)
- Policy Engine versionado
- PII, Secret, Prompt Injection, SSRF scanners
- Auto-redaction
- usage_records + budgets
- UsageRepo + BudgetRepo
- Alerts 80/90/100%
- OIDC (5 providers)
- SCIM 2.0
- Regional routing + GDPR
- Trace Store con rotación
- Request timeline
- X-Trace-Id header
- /livez /readyz probes
- Graceful shutdown
- Backup + DR tests
- Race condition en Cache.Get()
- Race condition en ValidateAPIKey()
- Snapshot() no copia mutex
- Migración 0005 sin expression index no-inmutable
- API keys SHA-256 hasheadas
- Idempotency-Key anti-duplicados
- SSRF protection metadata cloud
- PII auto-redaction
Gateway Pipeline, Events bus, Auth mandatory, Multi-tenancy, Cost tracking, OTel, Docker+K8s+Helm+Terraform.
Gateway con failover entre providers, Smart routing, Cache, Métricas, Dashboard.