The AI that runs your AI workforce.

---

The AI that runs your AI workforce. It learns your business, knows who's responsible for what, decides who handles a task, and answers for the outcome. The way a real CEO runs a company, not the way a chatbot answers a question.

CEO Agent is an orchestration layer that sits above your other AI agents and models. It learns your business, knows who's responsible for what, decides who handles a task, and answers for the outcome — the way a real CEO runs a company, not the way a chatbot answers a question.

You can rename her on install. The role doesn't change.

CEO Agent is built cloud-first, with mobile and (partial) desktop reach, for entrepreneurs and small-business owners who are tired of paying for a stack of mundane tools, tasks, and workflows one at a time — not as an enterprise platform play. It doesn't replace the AI tools already in someone's stack; it aggregates whatever they've already connected — Claude, ChatGPT, Grok, Gemini, their CRM, their email — under one executive layer that knows who owns what. CEO Agent is not competing with Hermes, OpenClaw, T3Agent, or any model provider — they're department heads and model options it coordinates, not rivals.

Status: working v1. The white-label identity layer, SDK, organizational model, setup wizard, CLI and local web chat interfaces, dispatch API, workflow execution engine, and safe skill-invocation proof of concept are built and tested. See Roadmap for what comes next.

---

Setup Wizard — Live Walkthrough

Five steps, about two minutes, re-runnable anytime.

Step 1 — Name your agent	Step 2 — Introduce yourself
Step 3 — Tell her about your business	Step 4 — Build her team
Step 5 — Cost mode and API key	Ready — org chart confirmed

First conversation, live:

Every department shown here is a real setup toggle: Finance, Operations (Hermes/COO), Technology, Marketing, People, and Legal. Skip the API key and CEO Agent still boots in routing-only mode until you connect a model provider.

Note: the setup-wizard screenshots above (steps 1–5, org chart summary) still match the current CLI. The chat screenshot does not — it predates the message hover actions (copy/edit/regenerate), the auto-resizing textarea, and citation rendering shipped since, and a planned navy/gold visual redesign will change it further. Flagging rather than retaking now, so it isn't retaken twice.

---

Install and Run It

Not yet published to npm — install from the repository.

git clone https://github.com/tradersurfer/ceo-agent.git
cd ceo-agent
npm install
npm run setup

Requires Node 18 or newer. The setup wizard will:

• Name your CEO Agent
• Ask about your business
• Let you activate whichever departments you need
• Choose flagship or efficient cost mode
• Save your OpenRouter key to a local, gitignored .env
• Launch straight into the chat interface

After setup, start the CLI again anytime:

npm start

Start the local-only web dashboard:

npm run web

In-Chat Commands

Command	Description
/org	Show your active org chart
/status	Runtime and agent status
/models	Resolved model assignments for both cost tiers
/cost	View or change cost mode (flagship or efficient)
/mode	View or change CEO Mode (conservative, aggressive, musk_mode)
/skills	List registered skills and their arguments
/attach <path>	Attach a local file to your next message
/help	Full command list
/exit	Quit

Address a department head directly with @department — for example, @legal draft an NDA clause. Anything else goes to CEO Agent directly.

Run a registered skill directly with /<skill> {json} or @<skill> {json} — for example, /format_currency {"amount": 42.5}. /skills lists every skill name and the arguments it accepts.

---

What Makes This Different

Most AI assistants answer one question at a time. CEO Agent runs the organization: she knows which department head owns which kind of work, delegates to them, and stays accountable for the result — while also answering directly when a task does not need to go anywhere else.

Capability	Description
Executive orchestration	Owns the top-level decision layer — decides what gets done, in what order, and by whom
Multi-agent task routing	Reads incoming work, identifies the right department, and delegates to the appropriate head
Model-agnostic operation	Resolves current models live through OpenRouter rather than hardcoding stale model identifiers
Cost-aware execution	Supports flagship and efficient model tiers, prompt caching, and real token-usage reporting
Workflow execution	Runs dependency chains, conditions, delays, retries, and registered bridge executors
Controlled skills	Provides a narrow, reviewed skill registry and executor pattern rather than arbitrary script execution
Business-vocabulary interface	Speaks in business, department, and employee terms instead of agent and prompt jargon

---

Supported Models

CEO Agent is model-agnostic. It resolves the current best available models live from the OpenRouter catalog into two cost tiers per role (flagship / efficient). No model IDs are hardcoded — they update automatically as providers release new versions.

Role	Provider	Icon
Claude	Anthropic
GPT	OpenAI
Gemini	Google
Grok	xAI
Codex	OpenAI (coding)

Use /models in chat to see the live resolved assignments for your current cost mode.

---

Architecture

                         CEO AGENT
                    Chief Intelligence
                     & Orchestration
                           │
       ┌──────────┬────────┼────────┬──────────┬──────────┐
       ▼          ▼        ▼        ▼          ▼          ▼
      CFO    HERMES/COO   CTO      CMO        CHRO       CLO
    Finance      Ops      Tech   Marketing    People     Legal
                                   │
                         ┌─────────┴──────────┐
                         ▼                    ▼
                    VP of Sales      Onboarding Comms

Seven departments, standard C-suite model. Hermes fills the COO seat directly (powered by NousResearch/hermes-agent), and Marketing includes a VP of Sales plus onboarding communications. The domain-specific Dispute Agent remains under examples/ as an optional reference implementation and is not activated in the default roster. Activate only the departments your business needs. See ARCHITECTURE.md for the full model.

---

Swarm Agents

CEO Agent coordinates a growing multi-agent swarm under the C-suite.

Agent	Status	Role	Icon
Hermes	Active	Chief Operating Officer / Ops bridge & workflow executor
OpenClaw	Coming soon	Swarm agent
T3Agent	Coming soon	Swarm agent

---

WORKSPACES™️

Coming soon — WORKSPACES™️ with Buzz by Block (relay Workspace connect).

This will enable secure, relay-based workspace connectivity so CEO Agent and the swarm can operate across distributed environments while keeping tenant isolation and auditability intact.

---

What's In This Repo

Path	What it is
IDENTITY.md	The CEO Agent's white-label identity template
departments-subagents/	C-suite identity documents and department-specific bridges
ceo-core/	Runtime configuration, departments, model routing, workflows, and skill execution
sdk/	Agent lifecycle, task routing, memory, permissions, prompts, and provider clients
organization/	The programmatic C-suite organizational model
registry/agent-registry.json	Canonical built-in agent roster and reporting structure
bin/setup.js	Interactive CLI setup wizard shown above
bin/chat.js	CLI chat with live model calls, cost switching, org chart, and status
app/	Local web dashboard and API routes for chat, configuration, status, org data, agents, and dispatch
departments-subagents/operations/hermes/	Hermes, the Operations department head and bridge, powered by NousResearch/hermes-agent
departments-subagents/marketing/sales-intake/	Marketing-owned sales intake bridge
departments-subagents/marketing/onboarding-comms/	Marketing-owned onboarding communications bridge
examples/dispute-agent/	Optional domain-specific reference implementation; not in the default roster
tests/	62 automated suites with 549 tests (npm test) covering workflows, bridges, skills, persistence, scheduling, rate limiting, health reporting, custom agents, runtime parity, user messages, department doctrine, multi-provider connections, CEO Modes, off_limits enforcement, the Hermes async lifecycle, Stripe webhook verification, and the web dashboard's components — plus 2 real-browser Playwright suites (npm run test:browser) for layout/interaction behavior jsdom can't exercise
ARCHITECTURE.md	The full department and agent product model
SECURITY.md	Current security posture, known limitations, and installer responsibilities
TASK_ROUTER.md, BEHAVIOR.md, ORGANIZATION-STRUCTURE.md	Reference behavior and routing specifications

---

Who This Is For

• A solo entrepreneur operating a local business
• A CFO or CEO who wants an executive AI layer over an existing technology stack
• A small-business owner coordinating multiple specialized AI tools or agents
• A developer building a white-label multi-agent installation

Available through both the CLI and the local-only web dashboard.

---

Roadmap

See docs/ROADMAP-MASTER.md for the standing 50-part, tiered plan. The checklist below is the short version — real shipped milestones, not the full backlog.

• White-label identity layer
• Core SDK for task routing, memory, permissions, and agent lifecycle
• Standard C-suite organizational model
• Department-head identity and prompt documents
• Operations and Marketing agent bridges, plus an optional Legal reference example
• Installer and setup wizard
• CLI chat with live model resolution and cost-mode switching
• Local web dashboard for chat, org chart, status, settings, and custom agents
• Dispatch API with authentication and rate limiting
• Workflow execution engine wired to registered bridge executors
• Safe skill-registry and executor proof of concept
• Cost and token optimization with dual-tier models, prompt caching, and usage visibility
• Security hardening and documented limitations
• Persistent workflow storage and audit log (Supabase-backed)
• Scheduler for delayed workflow steps
• Idempotency / duplicate-execution protection
• Multi-instance/shared rate limiting for hosted deployments (Supabase-backed)
• Health checks and observability
• Curated department skills: scope-creep detection, schema markup generation, content-quality analysis
• Hermes (COO / Ops) active in the default roster
• Real Hermes gateway execution — the bridge posts work to POST /v1/runs, not just validates and queues it
• BYNGE: direct per-provider connections (Anthropic, OpenAI, Google, xAI) alongside OpenRouter, with per-department model defaults and a live resolved-model catalog
• CEO Modes (Conservative / Aggressive / Musk Mode) — presets over the existing escalation-assessment and quality-review thresholds
• Agent Performance Visibility / Activity feed — who ran what, cost, and outcome, read-only in the web dashboard
• off_limits real code enforcement at both SkillExecutor.run() and BaseBridge.validatePermissions() (ADR-010)
• Hermes async lifecycle — real polling, layered timeouts, waiting_for_approval never auto-resolved
• Stripe webhook signature verification at the route layer
• Draggable/resizable sidebar
• Swarm agents: OpenClaw & T3Agent
• WORKSPACES™️ — relay Workspace connect with Buzz by Block
• Marketplace listing

Some install-specific pieces remain configurable by design: production runtime URLs for bridges, persistent storage, scheduling, model credentials, and deployment infrastructure.

---

Security

The included web commands bind to 127.0.0.1 by default. The dispatch and chat APIs rate-limit by caller: an in-memory sliding window by default, and a Supabase-backed shared limiter automatically when SUPABASE_URL/SUPABASE_SERVICE_ROLE_KEY are configured, for multi-instance deployments. Hosted deployments still need TLS termination, authentication appropriate to their environment, and a reviewed secrets-management process.

See SECURITY.md before deploying beyond a single local machine.

---

License

MIT — see LICENSE. Copyright (c) 2026 JECI Group, LLC.

---

Contributing

Issues and pull requests are welcome. Please avoid committing credentials, tenant-specific business data, personal paths, or private installation details.