What is the NylonME memory engine?

A single-node memory engine for AI agents by nylon-memory, distributed as a Rust project with a nylon-sdk package on PyPI: it stores memories as woven filaments in a mesh graph rather than a hierarchy of chunks.

How does NylonME relate to Jev?

The repo description says working with Jev can bring higher decision accuracy, but the README excerpt never mentions Jev, TypeSafe, or System One, so the integration is claimed rather than documented — treat it as a stated benefit, not a detailed interface.

How is retrieval different from top-k similarity search?

Context resonance: activation starts from seed nodes and spreads proportionally to edge strength times context match times real-time tension, bounded by a global activation budget, so high-fan-out nodes cannot explode the traversal.

What are the benchmark results?

On the full LoCoMo corpus (1536 answerable QA), evidence recall@10 went from 47.1% lexical to 85.9% after the write-side integrity fix; per-category recall is multi-hop 82.3%, temporal 89.1%, single-hop 89.1%, and commonsense 58.7%.

How are memories represented inside the engine?

Each memory is a braid of filaments — fact, emotion, temporal, relation, confidence, and frequency — with weighted edges between nodes forming a mesh, and unused memories decay along a tension forgetting curve instead of accumulating forever.

Is the API stable, and how new is the project?

The README labels the status Phase 2, active development, and warns that APIs may still evolve. Metadata: Apache-2.0, Rust, 52 stars, 7 forks, created 2026-08-04, last pushed 2026-10-16.

How do I install or try NylonME?

The badges link a nylon-sdk package on PyPI and releases on GitHub, and the README points to nylonme.com; the digest truncates inside the benchmark section, so setup commands, requirements, and the Chinese README contents are not in the excerpt.

Jev directory

CommunityJev Repos & SDKs

NylonME

A single-node memory engine for AI agents in Rust and Python: memories are woven from fact, emotion, temporal, relation, confidence, and frequency filaments into a mesh graph, retrieved by context resonance, and claimed to improve decision accuracy with Jev.

See highlights Open github.com

Stars: 52
Language: Rust
License: APACHE-2.0
Reading time: 3 min
Published by: Community
Author: nylon-memory
Last verified: Oct 2026

nylon-memory/nylonme

Organization repository on GitHub

View on GitHub

A Memory Engine for AI Agents,Working with Jev can bring higher decision accuracy . Nylon models a memory as a woven body of multiple strands of "threads" (facts/emotions/timing/relationships/beliefs/frequencies), with weighted edges between memory nodes forming a network graph rather than a hierarchical tree.

Forks: 7
Last pushed: Updated Oct 2026

Repo stats from the GitHub API, cached Oct 2026.

Tagscommunity github rust memory agents

Works forJev for Ranking

Used inJev for Software & DevTools

Languages: Rust Python
Capabilities: Open sourceSelf-hosted

Highlights

README: single-node memory engine for AI agents: memories are braids of filaments — fact, emotion, temporal, relation, confidence, frequency — not flat chunks.
Context resonance, not top-k: activation spreads from seed nodes by edge strength, context match, and real-time tension, bounded by a global budget.
LoCoMo evidence recall@10 on the full 10-session corpus climbed from a 47.1% lexical baseline to 85.9% after a write-side integrity fix (issue #1).
Per-category recall: multi-hop 82.3%, temporal 89.1%, single-hop 89.1%, commonsense 58.7%; end-to-end QA reached 82.9% on the paper protocol.
Apache-2.0 Rust repo in Phase 2 with APIs still evolving: 52 stars, 7 forks, created 2026-08-04, pushed 2026-10-16; a nylon-sdk package is on PyPI.

Watch out

README reviewed 2026-10-17 and truncated mid-sentence in the end-to-end QA section, so the Jev integration is not documented in the excerpt: the only link is the repo description's claim that working with Jev can bring higher decision accuracy. All recall and QA figures are the project's own LoCoMo runs, including the paired flips of +71 and −31, and no independent replication appears.

Questions about this entry

Short answers drawn from the linked project or page.

Reactions & coverage

Posts, threads, and videos about this entry from around the web.

Hacker News: jevmem keeps Claude Code project memory by asking Jev which turns are worth saving.

avinashjetwani61 pointsJevmem – automatic project memory for Claude Code, built on JevRead the thread on Hacker News

From the Jev guides

Original write-ups on what this entry is for.

The Open Reimplementations: Local Models That Speak System One

Sep 22, 2026

Closed weights have not stopped the ecosystem from rebuilding the shape in the open. API-compatible servers, retrained heads on small models, and inference-time tricks that skip generation entirely.

Jev Repos & SDKs#open-models#research#local

Read article

From the Jev AI model community

Posts from builders shipping with Jev right now.

Follow @typesafeai