The open-source Rust engine combines vector search, graph relationships and self-learning memory, now backed by Cognitum One commercial licensing, enterprise support and deployment services.

SAN FRANCISCO, CA / ACCESS Newswire / October 20, 2026 / AI agents forget. Each session starts from scratch, context is lost between tasks, and most vector databases store embeddings without recording how ideas connect or which answers proved useful. Cognitum One™, Inc. (cognitum.one) today launched RuVector™ commercially, adding licensing, enterprise support and deployment services to the open-source project. RuVector is an open-source Rust engine that combines vector search, graph relationships and self-learning memory for AI agents. It runs in PostgreSQL, in the browser and at the edge, so memory can sit wherever an organization’s data and agents already are.

RuVector has a large developer base behind it: more than 500,000 downloads a month across its npm and crates.io packages as of September 30, 2026, plus more than 4,400 GitHub stars and 580 forks. Within Cognitum One’s platform, the roles are distinct. Ruflo coordinates teams of agents, ruOS gives them governed computers, and RuVector is what they remember.

“Agents are only as good as what they remember,” said Reuven “rUv” Cohen, Founder and Chief AI Officer of Cognitum One and creator of RuVector. “Most vector stores can tell an agent what looks similar. RuVector also tells it how things are connected and which answers worked before, in one Rust engine that runs the same way on a server, inside Postgres, in a browser tab, or on a small device at the edge.”

Explore RuVector: https://cognitum.one/ruvector
Get started: npm install ruvector

What RuVector Does

  • Search and relationships in one engine. Fast vector search (HNSW indexing with SIMD acceleration, sub-millisecond queries) works alongside a Cypher-compatible knowledge graph. An agent can find similar material and follow how facts relate without a second database.
  • Improves retrieval from explicitly recorded feedback. When teams record which results helped, SONA updates lightweight LoRA adapters, and EWC++ consolidation protects earlier learning. Reads alone never change learned weights. Adapter learning is separate from index tuning and ranking settings, which teams control directly.
  • Works inside PostgreSQL. The ruvector-postgres extension is pgvector-compatible and adds 143 SQL functions for vector, graph and hybrid search inside an existing database. Teams should confirm compatibility for their specific pgvector workloads; full migration details are in the technical guide.
  • Signed, portable memory. The RVF file format packages vectors, indexes and metadata in a single file with copy-on-write branching and SHAKE256 witness chains, a tamper-evident record of every change. The same memory runs on servers, in WebAssembly in the browser and on edge devices, and running locally avoids hosted vector-service fees.

Why RuVector Stands Apart

RuVector extends conventional retrieval-augmented generation (RAG) by combining similarity search with graph relationships, persistent agent memory and learning from explicitly recorded feedback. It brings these capabilities together in one Rust engine for AI agents, RAG and intelligence workflows.

  • It learns instead of just storing. Traditional vector databases are passive storage. RuVector adds a graph neural network layer between initial search and final results that strengthens frequently used paths, and Cognitum One reports accuracy gains of 10 to 30 percent from usage feedback. SONA applies continuous LoRA micro-updates on a standard CPU, with Elastic Weight Consolidation protecting earlier learning from catastrophic forgetting.
  • A tech stack in a box. One engine replaces several systems: HNSW vector search, a graph engine with Cypher queries and hyperedges, and ruvllm, a built-in local runtime that runs language models and generates embeddings on-device with CUDA and Metal acceleration. An RVF file can bundle data, models and graph structures into a single artifact that boots a microkernel in milliseconds, making RuVector work almost like an operating system for agents at the edge.
  • Performance built in Rust. Production Rust with SIMD (AVX-512 and NEON) delivers search latency as low as 61 microseconds, which Cognitum One’s benchmarks put at 30 to 800 times faster than typical vector databases, and more than 52,000 inserts per second on commodity hardware.
  • A fraction of the memory. RuVector needs about 200 MB of memory per million vectors, compared with the 1 to 3 GB typical of other vector databases. Automated RaBitQ and product quantization compress data 2 to 32 times, so demanding workloads run on low-cost CPUs without GPUs.
  • Zero-config, everywhere. A single npm install ruvector detects the system and uses native Rust bindings, or falls back to WebAssembly for browsers and iOS. The PostgreSQL extension brings the same self-learning search to existing relational workflows.

In Practice

Illustrative example. A customer support agent receives a ticket about a login failure. RuVector retrieves a previous resolution for a similar ticket, follows its graph link to the product version it applied to, and confirms the fix still matches. When the support lead marks the answer as helpful, that feedback is recorded, and future retrieval for similar tickets improves across sessions.

“Enterprises want agents that remember their business, not just the last conversation, and they need to know where that memory lives and who changed it,” said D. Christopher Keil, Founding CEO of Cognitum One. “RuVector gives customers that control. It runs in the Postgres they already trust, keeps memory on their infrastructure, and makes every change auditable. Teams can start with a single install and grow into memory that is shared across agents, systems and sites.”

Availability

RuVector is available today. The core project remains open source under the MIT license. A Cognitum One commercial license adds a contract with enterprise support, deployment and configuration services, upgrade and maintenance commitments, and deployment with Ruflo™ and ruOS™ on infrastructure the customer controls. Model and infrastructure costs still apply.

Cognitum One recommends starting with one memory store and one agent, measuring retrieval quality on real queries, and adding graph, feedback learning and RVF features as each one earns its place. Installation commands, benchmarks and the full feature inventory are in the RuVector technical guide: github.com/ruvnet/ruvector.

About Cognitum One

Cognitum One is an artificial intelligence company building a sovereign agentic platform that orchestrates AI across the entire compute plane, from the public cloud to on-premises systems to the low-power edge, placing every workload where performance, cost, and security require. The platform is private by default, with real-time self-learning memory, distributed swarm coordination, contactless perception, cryptographic attestation, and native MCP integration, plus SDKs for Rust, Node.js, and Python. It is built for enterprises operating under data sovereignty and regulatory constraints. The company is led by Founder and Chief AI Officer Reuven Cohen, creator of the widely used open-source AI projects RuView™, Ruflo, and RuVector, and Founding CEO D. Christopher Keil. To learn more, visit cognitum.one.

PostgreSQL is a trademark of the PostgreSQL Community Association of Canada. Cognitum One is not affiliated with or endorsed by the PostgreSQL Community Association of Canada. Star and fork figures are from the public ruvnet/ruvector GitHub repository as of October 19, 2026. The download figure is the monthly total across RuVector’s npm and crates.io packages as of September 30, 2026. Performance, memory and accuracy figures are from Cognitum One’s internal benchmarks and depend on hardware, dataset size, recall settings and configuration.

Media Contact
Nick Ruest
Chief of Staff, Cognitum One, Inc.
nick@cognitum.one
(416) 477-3336

# # #

SOURCE: Cognitum One

View the original press release on ACCESS Newswire

About The Author
Cognitum One