Home/Special Projects/Court of Neri — Private Multi-Agent AI Operations Environment

Court of Neri — Private Multi-Agent AI Operations Environment

A local-first operating environment that routes real work across specialized agents, local models, tools, diagnostics, backups, and recoverable project infrastructure.

Private AI operationsApple Silicon + NVMeSpecialized agent routingOperational
Architecture diagram showing work routed through an OpenClaw gateway to specialized agents, local models, tools, data, diagnostics, and backup systems
Illustrative architecture map. The public diagram describes the operating logic without exposing private data, credentials, or implementation details.

The Court of Neri was not created as a product. It is Trident’s internal operating environment: a local-first AI workshop that routes different kinds of work to specialized agents, models, tools, and project contexts while remaining inspectable and recoverable.

The operating problem

Standalone AI chats fragment context and encourage repeated setup. Coding, market analysis, debugging, research, visual work, and operations do not benefit from identical model settings or identical system instructions. A useful environment needs role separation, durable workspaces, controlled tools, and an operational recovery path.

The constraints

01

Local first

Substantial work should be possible on the studio’s own Apple Silicon hardware and NVMe-backed storage.

02

Specialized roles

General assistance, long-context coding, action execution, debugging, vision, and market analysis require different behavior.

03

Memory discipline

Model residency, context length, parallelism, and storage must fit a finite unified-memory budget.

04

Recoverability

Configuration, secrets, health checks, backups, and repair steps must be visible enough to survive failure.

The system

An OpenClaw gateway provides sessions, routing, workspaces, and tools. Specialized agents are mapped to local models through Ollama, with configuration tuned for the studio’s M1 Ultra unified-memory environment. Heavy model and project data live on external NVMe storage. Diagnostic, backup, secret-loading, and recovery scripts keep the environment operational rather than theatrical.

Named roles include a general assistant, long-context coding agent, action runner, debugger, vision role, and Hermes—the quantitative trading analyst used alongside TridentHydra.

Operational design

The Court includes explicit health commands, configuration validation, local workspace rules, memory-pressure awareness, storage diagnostics, backup scripts, and recovery ladders. It treats the AI layer as infrastructure that must be maintained, not a magic interface that can be trusted blindly.

The Court’s value is not the mythology of the names. It is the operating discipline underneath them.

Relevant applications

Private business AI

Role-specific local assistants connected to carefully selected tools and internal knowledge.

Small-team operations

Research, documentation, coding, monitoring, and follow-up distributed across controlled agent roles.

Technical studios

Cross-platform environments that must support code, design, data, and unusual project work.

AI readiness

A practical reference for the storage, recovery, security, and maintenance questions often ignored during deployment.

What the project demonstrates

The Court demonstrates Trident’s capacity to architect and operate private agentic infrastructure: model routing, context design, local compute, storage planning, diagnostics, secrets, backup, and recovery in one working environment.

Status: operational internal infrastructure. It is presented as proof of capability, not as a packaged commercial product.

Project inquiry

Adapt the principle, not merely the interface.

Trident can discuss licensing, customization, private deployment, or a new system informed by the work.

Discuss the system