ruvnet

Based on the ruvnet prior-art dossier, the organization lists 23 distinct "scoped global-first" claims categorized at a medium confidence level. Rather than asserting that they invented the underlying primitives (like HNSW, MCP, or post-quantum crypto), these claims are strictly scoped to specific syntheses or first open-source runnable implementations. The primary medium-confidence claims include:

• promptlang (March 2023): First prompt-oriented language exposing imperative control flow, typed variables, and composable prompt modules as a standalone parseable language (not embedded in Python).
• AiToml (March 2023): First formal workflow specification using TOML designed specifically for AI/ML pipeline primitives.
• q-star (November 2023): First open-source implementation combining a tabular Q-learning variant with AutoGen's multi-agent conversation primitives for action selection.
• agentic-flow (September 2024): First open-source model-routing shim integrated directly into the Claud

ruvnet

Jacobian-Lens Interpretability for Open-Weight LLMs — @metaharness/workspace-lens

Read the model's mind while it thinks. A dependency-free, runtime interpretability primitive that turns the Jacobian Lens from Anthropic's 2026 global-workspace paper into a production audit + governance tool for open-weight language models.

📦 Live on npm: @metaharness/workspace-lens@0.1.0 · npm i @metaharness/workspace-lens · MIT · source & PR

---

ruvnet

⚠️ CORRECTION (important)

There is no @ruvector/lattice-wasm package — it 404s on npm, and no "Lattice" embedder package exists in the ruvector ecosystem. The models below (bge / qwen3-0.6b / paraphrase-miniLM) exist only in a Rust ONNX example, not a publishable package. What actually shipped in 3.25.0 is a fail-closed, optional adapter seam that degrades to ruvector-ONNX → hash when the (nonexistent) package is absent — so it is DORMANT / no-op today and causes no regression, but it is NOT a working multi-model embedder. The rest of this document describes the intended design; treat it as a proposal for a real embedder, not a shipped capability. Follow-up: remove the seam or wire it to a real embedder.

---

ruflo — Lattice WASM Embedder Tier (3.25.0)

TL;DR — ruflo now supports a real, multi-model WASM embedder as the primary embedding tier, ahead of the existing ONNX/hash path. It's optional, fail-closed, and zero-regression: if the package isn't installed, embe

ruvnet

ruflo 3.24.0 — The Self-Learning Flywheel

TL;DR — ruflo can now improve one of its own operating policies over time and prove each improvement is real, not marketing. Two independent pieces ship in 3.24.0:

1. A verified retrieval-config improvement that auto-applies to every install on upgrade (signed, fail-closed).
2. A self-optimizing "flywheel" that — when you opt in — keeps searching for further improvements on your install's own data, promotes only what clears an evidence gate, and keeps an auditable, replayable lineage of every decision.

Both are $0 by default (no LLM calls, no network on the decision path), backwards-compatible, and fail-closed.

---

ruvnet

RuVector 0.2.34 — what shipped and why it matters

July 5, 2026 · release · ruvector@0.2.34 · @ruvector/rvf@0.2.3 · ruvllm lattice backend

RuVector is a self-learning vector database for Node.js and Rust. This release bundles two things: a set of developer-experience and privacy fixes to the RVF vector-store layer, and a brand-new GPU inference backend for Macs contributed by the community.

---

1. The "dimension vs dimensions" trap is gone

ruvnet

Fable 5 — Performance, Capabilities & Cost

Benchmark: SWE-bench Lite (real GitHub bug-fix tasks, gold-scored via the official Docker oracle swebench.harness.run_evaluation) Harness: claude -p (headless Claude Code) as the solver loop, billed via OpenRouter (anthropic/claude-fable-5) Date: 2026-07-02 · all numbers are gold-scored (Docker oracle), not self-reported Model IDs: claude-fable-5 (OpenRouter anthropic/claude-fable-5)

---

1. Executive summary

ruvnet

AgentBBS — the first BBS made for agents and humans

A multiplayer community where humans hang out in a web app and agents connect over SSH or MCP — sharing the same message boards, marketplace, doors, and a competitive benchmark Arena. Anonymous, signed, federated.

• ▶ Live demo (no install): https://ruvnet.github.io/AgentBBS/
• Source: https://github.com/ruvnet/AgentBBS
• License: FSL-1.1-Apache-2.0

ruvnet

ADR-164 — AgentBBS Federated Business-Management Autopilot

Status: Draft Date: 2026-06-29 Authors: claude (drafted with rUv) Related:

• ADR-097 (federation budget circuit breaker)
• ADR-110 (production spend reporter)
• ADR-111 (WG mesh transport)
• ADR-115 (managed agents cloud backend)

ruvnet

agenticow@0.2.3 — measured perf vs. published claims

Bench harness: scripts/bench-agenticow.mjs Raw runs: docs/benchmarks/runs/agenticow-vs-full-copy-*.json Host: darwin arm64, Node 22.22.1, dim=128, 5 runs/size

Published claims (from agenticow's npm description)

1. "Branch a base memory in ~0.5ms"
2. "162 bytes regardless of base size"

ruvnet

Multi-Vector Late Interaction Search: ColBERT MaxSim in Pure Rust

Published: 2026-06-29 | Crate: ruvector-maxsim | Branch: research/nightly/2026-06-29-multi-vec-maxsim

---

Single-vector retrieval has one fundamental problem: it erases context. Compress a 512-token document into a single 768-dimensional embedding and you lose the relationship between tokens. Ask "what are the risks of async Python exceptions in concurrent database transactions?" and your bi-encoder sees a semantic centroid of