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	13926
title	Graph-independent deployment-state bridge for inspect_deployment
state	Closed
labels	enhancementaiarchitecturemodel-experience
assignees	neo-gpt
createdAt	Jun 23, 2026, 2:55 PM
updatedAt	3:22 PM
githubUrl	https://github.com/neomjs/neo/issues/13926
author	neo-gpt
commentsCount	1
parentIssue	null
subIssues	[]
subIssuesCompleted	0
subIssuesTotal	0
contentTrust
projected
quarantined	0
signals	[]
blockedBy	[]
blocking	[]
closedAt	Jun 23, 2026, 9:17 PM

Graph-independent deployment-state bridge for inspect_deployment

Closed v13.1.0/archive-v13-1-0-chunk-6 enhancementaiarchitecturemodel-experience

neo-gpt commented on Jun 23, 2026, 2:55 PM

Parent: #13860

Related: #13914, #13920, #13923, #13924, #13874, ADR-0025, ADR-0026.

Problem

#13914 correctly targets a public inspect_deployment MCP surface, but its current shape still leans too hard on Memory Core graph readout. That is not sufficient for the live cloud failure class: the model runner can pin CPU, Memory Core can be unhealthy, and graph publication/readout can lag or be unavailable exactly when deployment proof is needed.

Only KB and MC are public MCP surfaces. The Docker socket must stay off both of them. The missing substrate is a graph-independent bridge from the internal orchestrator daemon to those public tools.

Scope

Add a model-free, graph-independent deployment-state bridge that #13914 can consume:

orchestrator remains the only holder of Docker/runtime access (#13920);
KB/MC get read-only MCP tools, not a socket mount and not a write actuator;
latest diagnostics/recovery input is readable without an embedding/model call and without relying 100% on graph availability;
MC-down cases still have a public path through KB if KB is healthy;
payloads are bounded and allowlisted to Neo deployment services only.

Acceptance Criteria

No Docker socket mount on KB or MC.
No public orchestrator route.
inspect_deployment can read latest daemon diagnostics/recovery input through a model-free bridge, not only graph records.
The bridge supports MC-unhealthy / graph-unavailable failure mode by allowing a KB-hosted read path when KB is healthy.
Graph remains an archive/SSOT when available, but is not the only proof surface.
[L3-deferred - operator deployment handoff needed] Live-smoke fixture covers a wedged model runner plus unhealthy memory container and proves the state through the public MCP tool path.

Contract Ledger

Contract	Owner	Consumer	Required behavior	Fallback / failure mode	Evidence
Internal diagnosis snapshot	orchestrator daemon	bridge writer	bounded, allowlisted latest deployment diagnosis; no model call	stale/unavailable is explicit, not silent	unit + fixture
Graph-independent bridge payload	bridge substrate	KB + MC tools	read-only snapshot with size cap, timestamp, source, and service allowlist	MC unhealthy still leaves KB path if KB is healthy	live smoke
Public KB read tool	KB MCP	inspect_deployment / operators	read current snapshot without Docker/socket/shell authority	returns structured unavailable/stale result	MCP schema + test
Public MC read tool	MC MCP	inspect_deployment / graph archive	read graph/archive when healthy, but not required for current proof	graph unavailable does not mask KB snapshot	MCP schema + test
Deployment wiring	private deployment config	stack runtime	internal-only bridge mount/env; no public route	missing config fails closed and logs the exact migration key	private MR + short migration note

Non-negotiables: no public orchestrator or daemon route; no Docker socket on KB or MC; no shell/exec/write tool; bounded allowlisted payloads; model-free read path; graph remains archive/SSOT when available, not the only proof surface.

Security invariant

Public surfaces remain KB and MC only. Runtime authority remains internal to the orchestrator holder and is never exposed as an arbitrary container operation, shell, exec, or public write tool.