Release v2.0.6 — TurboQuant, Brain Cognition, Resend Email, 30+ Crate Publishes

[ruvnet]

Release v2.0.6

412 commits since v2.0.5. This release spans brain server cognitive enhancements, TurboQuant KV cache compression for ruvLLM, Resend email integration, and a full crates.io publish cycle.

Highlights

TurboQuant KV Cache Compression (PR #297)

• TurboQuant compressor: Data-oblivious ~3.5-bit compression with provable geometry preservation (ICLR 2026)
• Three-tier KV cache: FP16 hot tier + TurboQuant cold tier with auto-migration
• Optimized inner product: Rotated-domain computation skips inverse Hadamard (~2x faster attention)
• Embedding store: Compressed vector search compatible with HNSW
• Benchmarks: Criterion suite covering all operations, dimension scaling, memory efficiency
• 18 passing tests, 4 bit-width configs (2.5/3.0/3.5/4.0)

Brain Server Cognitive Loop (PR #296, merged)

• Enhanced cognitive tick: 60s lightweight + 5min full cycle (SONA + symbolic + GWT + curiosity)
• Gist publisher: Autonomous publishing of novel discoveries with novelty gates
• Symbolic reasoning: Cross-domain inference, expanded rule engine
• Google Chat bot: Cards V2 handler with text fallback
• Resend email: Inbound webhook, subscribe/unsubscribe, pixel tracking (ADR-125/126)

MoE & Quantization (earlier in cycle)

• ADR-090: Ultra-low-bit QAT with Pi-quantization SIMD kernels
• ADR-091: INT8 CNN quantization for edge inference
• ADR-092: MoE memory-aware routing with P1-P4 optimizations
• TurboQuant (ADR via PR feat(ruvllm): TurboQuant KV cache & vector compression #297): Two-stage PolarQuant + QJL compression

Brain Platform

• 18 MCP tools (cognitive status, LoRA, training, pipeline)
• AGI self-optimization: self-reflection, inference, adaptive SONA
• Sparsified MinCut, Hopfield recall, LoRA auto-submit
• Common Crawl adapter, PubMed pipeline, PiQ3 quantization
• Partition caching, SSE session improvements

Infrastructure & Ecosystem

• MinCut performance: flat capacity matrix + allocation reuse (10-30% faster)
• Dynamic MinCut (ADR-124), source-anchored canonical implementation (ADR-117)
• SONA state persistence (loadState/saveState), trajectory fix
• RuVix cognition kernel + CLI + PBFT consensus
• Neural Trader crates + WASM bindings
• DrAgnes + Common Crawl WET + Gemini grounding agents
• ruvector-sparsifier: dynamic spectral graph sparsification
• ruvector-dither, thermorust, ruvector-robotics crates

Crates Published to crates.io

Crate	Version
ruvector-math	2.0.6
ruvector-filter	2.0.6
ruvector-metrics	2.0.6
ruvector-profiler	2.0.6
ruvector-snapshot	2.0.6
ruvector-collections	2.0.6
ruvector-graph	2.0.6
ruvector-cluster	2.0.6
ruvector-mincut	2.0.6
ruvector-solver	2.0.6
ruvector-gnn	2.0.6
ruvector-attention	2.0.6
ruvector-coherence	2.0.6
ruvector-sona	0.1.9
ruvector-replication	2.0.6
ruvector-raft	2.0.6
ruvector-dag	2.0.6
ruvector-sparse-inference	2.0.6
ruvector-temporal-tensor	2.0.6
ruvector-domain-expansion	2.0.6
ruvector-nervous-system	2.0.6
ruvector-cognitive-container	2.0.6
ruvector-router-core	2.0.6
ruvector-router-ffi	2.0.6
ruvector-router-wasm	2.0.6
ruvector-router-cli	2.0.6
ruvector-tiny-dancer-core	2.0.6
ruvector-graph-node	2.0.6
ruvector-graph-transformer-node	2.0.6
ruvector-mincut-node	2.0.6
ruvector-solver-node	2.0.6
ruvector-gnn-node	2.0.6
ruvector-attention-node	2.0.6
ruvector-tiny-dancer-node	2.0.6
ruvector-graph-wasm	2.0.6
ruvector-mincut-wasm	2.0.6
ruvector-solver-wasm	2.0.6
ruvector-gnn-wasm	2.0.6
ruvector-exotic-wasm	2.0.6
ruvector-attention-wasm	2.0.6

Release Checklist

• All crates published to crates.io at v2.0.6
• PR feat(brain): Resend email integration with pixel tracking #296 merged (brain + email integration)
• PR feat(ruvllm): TurboQuant KV cache & vector compression #297 auto-merge enabled (TurboQuant)
• Issue feat(ruvllm): TurboQuant KV Cache & Vector Compression — Full Implementation Plan #298 created (TurboQuant implementation plan)
• Tag v2.0.6 on main after PR feat(ruvllm): TurboQuant KV cache & vector compression #297 merges
• Create GitHub Release with binaries
• Update CHANGELOG.md
• Verify crates.io pages render correctly

Related PRs & Issues

• PR feat(brain): Resend email integration with pixel tracking #296: feat(brain): Resend email integration with pixel tracking — Merged
• PR feat(ruvllm): TurboQuant KV cache & vector compression #297: feat(ruvllm): TurboQuant KV cache & vector compression — Auto-merge pending
• Issue feat(ruvllm): TurboQuant KV Cache & Vector Compression — Full Implementation Plan #298: TurboQuant full implementation plan

[ruvnet]

Release v2.1.0 Published

This release has been superseded by v2.1.0, which includes everything from v2.0.6 plus 13 additional SOTA modules.

TurboQuant Status: ✅ Complete

• 1,483 lines in `turbo_quant.rs` + 1,766 lines in `kv_cache.rs`
• 2/3/4/8-bit asymmetric quantization, H2O/PyramidKV eviction
• Asymmetric inner product on quantized data
• 18 tests, all passing
• Published: `ruvllm@2.1.0` (crates.io), `@ruvector/ruvllm@2.5.4` (npm)

v2.1.0 Additional Modules

• Hybrid Search (RRF), Graph RAG, DiskANN, ColBERT, Matryoshka, OPQ, LSM Compaction
• FlashAttention-3, MLA, KV-Cache Compression, Mamba SSM, Speculative Decoding
• GraphMAE

Published Packages (v2.1.0)

6 Rust crates + 6 npm packages — see release notes

This issue can be closed in favor of the v2.1.0 release.