Skip to content

NETRA Product Roadmap

This document outlines the phased engineering milestones for NETRA. Milestones follow standard delivery stages: Complete and Scheduled.


Phase Milestones Overview

Phase Target Release Milestone Name Status Key Deliverables
Phase 1 v0.1.0 Foundation & Ingress Pipeline Complete Java 25 LTS baseline, Spring Boot 4.1.1, Docker Compose stack, canonical data seeder, strict x-netra-* telemetry contract, OpenAPI 3.1 specification.
Phase 2 v0.2.0 Virtual Key Auth & Atomic Admission Complete Multi-tenant auth, model allowlists, atomic Redis 8.0 Lua sliding-window rate limiting (0.00% over-admission), monthly budget caps, request audit logs.
Phase 3 v0.3.0 Unbuffered Streaming SSE Relay Complete Token-by-token pass-through streaming over Server-Sent Events on Java 25 Virtual Threads, terminal chunk usage extraction, mid-stream disconnect handling.
Phase 4 v0.4.0 Multi-Provider Routing & Auto Classifier Complete Ordered fallback chains, exponential backoff retries (3 attempts), fast-failover across providers, two-stage prompt difficulty classifier for auto tier (~72.4% cost savings).
Phase 5 v0.5.0 Tenant-Isolated Semantic Vector Cache Complete In-JVM dense embeddings via quantized all-MiniLM-L6-v2 (384-dim), PostgreSQL 18 pgvector HNSW index, strict cryptographic tenant isolation, refusal suppression, streaming SSE cache replay ($0 upstream cost).
Phase 6 v0.6.0 Ops Console & Enterprise Verification Suite Complete React 19 / Vite 6 Ops Console (console/), OpenRouter-grade analytics KPIs, single-denomination forex toggle (USD/INR), live SSE request stream, virtual key manager, prompt playground, multi-tier PaaS deployment blueprints (Railway, Render, Fly.io, Vercel, Cloud VPS), and v2 verification suite (scripts/v2/).
Phase 7 v0.7.0 Enterprise Resilience, Two-Tier Cache & High-Load Hardening Complete Two-tier cache hierarchy (Tier-1 exact SHA-256 hash <0.5ms + Tier-2 pgvector semantic cache <5ms), standardized IETF RateLimit headers, Resilience4j dynamic circuit breakers, provider health telemetry (GET /admin/providers/health), small-model refusal auto-escalation engine, Java 25 Scoped Values request context propagation, decoupled batch database logging (10,000 bounded queue), and in-memory multi-tenant spend aggregation.
Phase 8 v0.8.0 Standard Observability & Prompt Cache Accounting Scheduled Vendor-neutral OpenTelemetry GenAI semantic conventions, provider-side prompt caching discount tracking (cached_tokens), dual-model comparison playground, and stream latency benchmarking.
Phase 9 v0.9.0 Self-Hosting Hardening & Webhook Alerts Scheduled Production Kubernetes Helm charts, single-container zero-config standalone mode, budget threshold webhook alerts (Slack/custom HTTP), and Redis Sentinel failover.
Phase 10 v1.0.0 Production LTS Baseline Scheduled Formal 1.0 API stability contract, backwards-compatibility guarantees, native provider adapters (Anthropic Messages, Vertex, DeepSeek), and production LTS hardening.