Supporting Case Study #103D Multi-Agent Workstation35 Specialist PersonasLocal Brain Notes Graph
Agents Office: Local Multi-Agent Business & Operations Workstation
A local 3D multi-agent workstation where specialist AI personas route operational tasks, index user-owned Markdown notes, coordinate multi-agent teams, run scheduled calendar routines, and file deliverables with strict human approval gates.
Agent Roster35 Roles / 6 Pods
Knowledge LayerLocal Brain Graph
InterfaceThree.js 3D Office
Safety ModeHuman Approval Gate
Three.js 3D Isometric Office & Department Pods
Expand ProofThe Workstation Problem
Why Linear Chat Interfaces Fail Complex Multi-Role Operations
Context Dilution in Single Chat Threads
Mixing market research, copy drafting, SEO reviews, and technical debugging into one long chat thread clutters context windows and degrades prompt performance.
Unchecked Outbound Action Risks
Giving agents direct access to external connectors without visual human review gates risks accidental external communication, erroneous payments, or unauthorized data changes.
Knowledge Isolated from Local Notes
Cloud AI platforms lock conversation history inside proprietary databases, disconnected from the user's local workspace and existing business documentation.
Lack of Scheduled Operational Cadence
Growth and operations teams require recurring checks (weekly audits, daily queue updates). Standard chat bots do not operate on a calendar schedule.
The Architecture
Agents Office: Grounded Local Multi-Agent Workstation
I built Agents Office as a local-first workstation that models AI collaboration as a physical office. It organizes 35 specialized personas across 6 departments, indexes local notes into a searchable semantic graph, coordinates team deliverables, and runs recurring operations on a calendar with human approval checkpoints.
3D Spatial Office
Three.js isometric canvas visually represents agent desks, current workload states, and routing queues.
Local Brain Graph
Grounds reasoning directly in user-owned Markdown notes, maintaining local data ownership and privacy.
Approval Gates
Requires interactive human approval before any external write, financial transaction, or email dispatch.
Task Execution Lifecycle
How It Works: From Local Brain to Verified Deliverable
1
Brain Index
Scans local Markdown notes into context graph.
2
Task Routing
Routes request to appropriate department pod.
3
Team Collab
Lead coordinates researcher and copywriter.
4
Routines
Calendar schedules recurring operational tasks.
5
Approval Gate
Human must explicitly sign off on writes.
6
Deliverable
Stores output note back in user's local brain.
Agents Office Workstation State Machine
Node.js 20 ES Modules# Routing task to Marketing pod agent: Copywriter Lead
> Input: "Draft email sequence for high-intent GSC striking distance queries"
> Router: matched department = 'marketing', agent = 'copywriter', teammates = ['seo_reviewer']
> Context loaded from brain: 4 notes linked via tag #gsc_opportunities
⚠️ External Connector [AWS_SES_GATEWAY] requested: Awaiting User Approval [Approve/Reject]
Evidence Gallery
Visual Evidence & Workstation Screenshots
3D Workstation
3D Isometric Multi-Agent Office & Department Layout
Interactive Three.js 3D isometric office layout visualizing 6 department pods (Marketing, Operations, Analytics, Engineering, Strategy, Content) and their assigned agent workstations.
Click for proof
Brain & Roster
Local Brain Knowledge Graph & Specialist Agent Roster
Local knowledge graph connecting user markdown notes with a 35-agent specialist roster, mapping semantic relationships across projects, tools, and brand guidelines.
Click for proof
Routines Engine
Automated Operations Calendar & Routines Engine
Scheduled routines view tracking recurring growth operations: automated weekly Search Console reviews, daily ad spend audits, and social queue refills with explicit human approval gates.
Click for proof
Verification & Rigor
What These Engineering Artifacts Conclusively Prove
Visual Multi-Agent Organization
This proves: Replaces chaotic terminal scripts with a structured 3D office and 6 department pods, making agent routing and team coordination inspectable.
Why it matters: Allows non-engineers and team operators to observe AI workflows and collaborate safely with autonomous roles.
Local Brain Context Preservation
This proves:Indexes local Markdown notes and files directly, saving deliverables with full provenance back into the user's folder.
Why it matters: Keeps business knowledge private and durable without depending on external vector database subscriptions.
Workstation Modules
Core Workstation Capabilities
6 Department Pods & 35 AgentsAgent Swarm
Organized into Marketing, Growth, Operations, Content, Engineering, and Analytics with dedicated role descriptions.
Local Brain Knowledge GraphLocal-First
Indexes local Markdown notes and files to supply context to agents without uploading sensitive data to cloud vector databases.
Collaborative Team TasksOrchestration
Enables multi-agent collaboration where a lead agent coordinates teammates (researcher, drafter, reviewer) to produce deliverables.
Routines & Calendar SchedulingAutomation
Schedules recurring marketing and analysis routines on a visual calendar with automatic catch-up execution upon app launch.
Human-in-the-Loop Approval GatesGovernance
Outbound operations (sending emails, modifying ad spend, publishing pages) require explicit interactive human sign-off.
Claude Code MCP Connector DiscoveryTooling
Automatically discovers and connects Model Context Protocol tools with strict allow/deny permission policies.
Under The Hood
Technical Stack & Local Workstation Architecture
Runtime & Agent Engine
- Node.js 20 ES Modules: Lightweight, modular server executing agent routing, routine scheduling, and JSON state persistence.
- Claude Code MCP Integration: Discovers local Model Context Protocol tools with strict allow/deny permission policies.
- Offline Smoke Check: Validates clean startup and JSON integrity with
npm run check.
Frontend & Spatial Graphics
- Three.js 3D Canvas: Lightweight isometric 3D office rendered directly in the browser via esbuild bundle.
- D3 Force Layout: Dynamic interactive graph mapping relationships between notes, topics, and deliverables.
- Local Filesystem Brain: Direct read/write to Markdown notes with zero remote database synchronization.
Epistemic Rigor
Verified Repository Evidence (Zero Hallucinated Metrics)
[FACT] 35 Documented Roles: Defined in
office.agents.json across 6 department pods.[FACT] Beta Release Labeled: Source explicitly notes Beta status with human approval required for outbound actions.
[FACT] Local Notes Ground Truth: All context sourced from local Brain folder without cloud vector locks.
[FACT] Zero Fake Adoption Claims: Does not claim automated enterprise revenue or human replacement figures.
Commercial Impact
Business Value & Operator ROI
Visual Operations
Makes complex multi-agent workflows transparent and easy to track for business stakeholders.
Zero Data Leakage
Operates over local Markdown notes, protecting intellectual property and sensitive customer data.
Safe External Actions
Guarantees human approval before any real money is spent or external message is dispatched.
Scope & Transparent Boundaries
Engineering Limitations & Scope
Beta Product State: The repository explicitly states that Agents Office is in active Beta development and should be expected to have rough edges.
API Key Prerequisite: Requires an active Claude authentication session or Anthropic API key to execute intelligent reasoning.
Human Approval Mandate: The system enforces approval gates by design; it is not intended for unmonitored financial autonomy.
Ecosystem Position
Supporting Technical System in the Portfolio
Agents Office demonstrates product design for multi-agent systems, human-in-the-loop governance, and spatial UI development. It acts as the orchestration environment that connects specialized growth skills (SearchFoundry, TrafficSentinel, Marketing MCP) into a cohesive everyday workplace.
What This Demonstrates to Engineering Teams
Agents Office illustrates my ability to design intuitive human-computer interfaces for complex agentic workflows. It proves competence in 3D browser graphics (Three.js), local-first data indexing, and robust approval-state governance.
Available for 2026 Roles
Let's build repeatable growth engines.
I'm open to Digital Growth, Technical SEO, SaaS, AI Automation, and Analytics opportunities.