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.

Frontmatter

id	14559
title	KB release artifact: binary (fp16) vector encoding to cut the JSON-float text tax (low-pri, before v13.2)
state	Open
labels	enhancementai
assignees	neo-opus-grace
createdAt	12:03 AM
updatedAt	12:03 AM
githubUrl	https://github.com/neomjs/neo/issues/14559
author	neo-opus-grace
commentsCount	0
parentIssue	null
subIssues	[]
subIssuesCompleted	0
subIssuesTotal	0
contentTrust
projected
quarantined	0
signals	[]
blockedBy	[]
blocking	[]

KB release artifact: binary (fp16) vector encoding to cut the JSON-float text tax (low-pri, before v13.2)

Open Backlog/active-chunk-2 enhancementai

neo-opus-grace commented on 12:03 AM

Priority: LOW — look into before the next release, not now. fp16 alone buys years and the GitHub cap is soft (below); no urgency.

Context

v13.1.0's KB release artifact (chroma-neo-knowledge-base.zip, via ai/scripts/maintenance/uploadKnowledgeBase.mjs) hit 1009.91 MB — 49,283 chunks, dim 4096. ~97% is embedding vectors serialized as JSON-float TEXT; the source conversations are only ~28 MB. Growth driver = chunk count (tickets + PR conversations, intense).

Invariant (do not break)

The artifact ships raw vectors precisely so adopters never re-embed 49k×4k on boot. Any fix MUST preserve raw-vectors / no-re-embed. "Ship text, re-embed on download" is out.

Lever — binary encoding (the waste is encoding, not vectors)

Verified arithmetic: fp32 binary uncompressed = 770 MB (already < the current 1010 MB zip); fp16 = 385 MB.

fp16 binary ≈ 2.6× smaller, minor precision drift, no re-embed → the KISS win.
fp32 ≈ 1.4×, lossless. fp16-vs-fp32 is a recall-vs-size trade-off worth a quick measurement before committing.
Impl: export schema v1→v2 (packed/binary vectors), decode download-side, keep the meta.json drift-check.

Secondary (defer)

Shard past 2 GiB — GitHub allows 1000 assets/release, no total-size/bandwidth cap → sharding is a trivial runway; size never blocks. Safety net only.
Baseline scope — the real growth lever: does the shipped KB need the whole ever-growing dev-conversation corpus, or durable knowledge + a separate conversation sync? Product decision.

Refs: ai/config.mjs:372 (dim 4096), uploadKnowledgeBase.mjs, v13.1.0 cut. Analysis persisted; owner @neo-opus-grace.

Authored by Grace (@neo-opus-grace, Claude Opus 4.8).