What is the Neural Link?

The Neural Link is a bi-directional bridge that connects AI agents directly to the Neo.mjs runtime. It lets agents inspect the Scene Graph, component state, event listeners, computed styles, and DOM rectangles, and mutate the running application in real time.

Why is Neo.mjs called an Application Engine instead of a framework?

Neo.mjs maintains persistent application objects in a worker-backed Scene Graph instead of compiling application state away into ephemeral DOM nodes. That architecture enables multi-window orchestration, runtime permutation, and deep AI introspection.

What is Context Engineering?

Context Engineering shapes the information and tool environment around AI agents. Neo.mjs implements it through Knowledge Base, Memory Core, GitHub Workflow, and Neural Link MCP servers for frontier harnesses, plus a File System MCP server for internal Neo.ai.Agent local loops.

What is the Neo.mjs Agent OS?

The Neo.mjs Agent OS is the repository Brain: source code and services for Memory Core, Knowledge Base, Active Hybrid GraphRAG, DreamService, Golden Path synthesis, A2A coordination, and Neural Link tooling.

Neo.mjs - Self-Evolving Software Organism

Frontmatter

title	>-
author	tobiu
state	Merged
createdAt	Apr 15, 2026, 12:37 AM
updatedAt	Apr 15, 2026, 10:49 AM
closedAt	Apr 15, 2026, 10:49 AM
mergedAt	Apr 15, 2026, 10:49 AM
branches	`dev` ← `agent/9999-optimize-local-summarization`
url	https://github.com/neomjs/neo/pull/10019

Merged

tobiu commented on Apr 15, 2026, 12:37 AM

Description

Resolves #10021

This PR removes the lossy map-reduce bottleneck and hard slicing that was destroying local inference performance for Gemma4 on the M5 processors. The map-reduce loop resulted in 30-60 min runtimes for 17 sessions.

Architectural Changes

Removed map-reduce chunking from SessionService summarization.
Removed slice(-3) from DreamService REM loop.
Updated SessionSummarization tests.

tobiu commented on Apr 15, 2026, 10:46 AM

Input from Antigravity (Gemini 3.1 Pro):

✦ # PR Review Summary

Status: Approved

Thanks for putting this together! Great approach to stripping out the legacy map-reduce overhead. Sending the raw documents directly makes perfect sense since local models can now gracefully handle massive n_ctx capabilities via Flash Attention. This is ready to merge.

📊 Evaluation Metrics

[ARCH_ALIGNMENT]: 100 - Code successfully eliminates intermediate looping overhead, pushing the context parsing directly down to the LLM backend which fundamentally aligns with Neo's high-performance patterns.

[CONTENT_COMPLETENESS]: 90 - Testing explicitly asserts the map-reduce bypass by validating prompt count (toBe(1)) and length (> 20000).

[EXECUTION_QUALITY]: 95 - Removing slice boundaries enforces lossless abstraction. The <20s API duration test provides a strict contract against performance regressions.

[PRODUCTIVITY]: 100 - Directly addresses the 30-60 min latency bottleneck.

[IMPACT]: 80 - Solves a critical operational barrier for developers running the Agent OS locally.

[COMPLEXITY]: 30 - A targeted removal of complex code in favor of a simpler, native implementation.

[EFFORT_PROFILE]: Quick Win - High ROI for eliminating a massive local latency bottleneck by paring down logic.

🕸️ Context & Graph Linking

Target Epic / Issue ID: Resolves #10021

Related Graph Nodes: MemoryCore, SessionSummarization

🧠 Graph Ingestion Notes

[RETROSPECTIVE]: Removing artificial software abstraction (like Map-Reduce chunking) when the underlying foundational models natively support massive context windows yields immense performance multipliers. Always evaluate if the application layer algorithms are trying to solve capacity limits that the backend has already conquered.

📋 Required Actions

To proceed with merging, please address the following:

Code is clean. No further actions required.

LGTM. Ready for squash and merge into pending dev.