Platform

A command center for an office that runs itself — within the limits you set.

Everything the agents do is an event. The portal subscribes to all of them, so every page is live without refreshing.

Portal

Sixteen pages, one WebSocket.

Built with Next.js 15, React Three Fiber and zustand; charts and graphs are hand-drawn SVG, no heavy dependencies.

Command Center

Live KPIs, office status, approvals, active tasks, agent messages and spend.

Virtual Agent Office

3D office with five desks, a holographic orchestration table, LED wall and five camera modes.

Agents & chat

Profile, tools, memories, live sandbox log and a streaming chat with each agent.

Collaboration

SVG graph of agent-to-agent handoffs with the orchestrator as hub.

Tasks

Pipeline board, cancel, drawer with the messages tab.

AWS Accounts

Three-step onboarding wizard: scope → CloudFormation → verify.

Infrastructure

Environments with health probes and a per-account inventory viewer.

Deployments

Pre-checks, approval, execution lock, verification, rollback.

Monitoring

System metrics, environment health, incidents and the LED-wall feed.

Security

Findings, release gate, authorised targets, run audit.

Knowledge

Runbooks and documents with permission-aware retrieval.

Reports

Decks, models and PDFs produced by the agents.

Approvals

Pending and history, role-gated Approve / Reject with countdown.

Integrations · Settings · Audit

Connector status, MFA, users, budgets, retention, contact inbox, append-only audit.

Modes

DEMO=1 by default. DEMO=0 when you are ready.

In DEMO mode the LLM is a deterministic stub and every tool returns realistic, labelled output — the whole system runs end-to-end with zero cloud keys.

Honest labelling

Every simulated payload carries demo: true and the UI shows a DEMO chip. Buttons that cannot act are disabled with the reason.

Same code path

Routing, policy, approvals, collaboration bounds and audit behave identically in both modes — only the tool back-ends change.

Switching

Set DEMO=0, provide Bedrock (or Anthropic / OpenAI) credentials, AWS roles and Tailscale, then restart the stack.