Create, configure, train, and manage custom AI agents — each with its own expertise, memory, and API. No coding required. Clone from templates. Train with your data. Deploy anywhere.
ChatGPT doesn't know your sales process. Claude hasn't read your employee handbook. No off-the-shelf AI understands your specific customer segments, pricing model, or operational quirks. That's the gap.
Train them on your documents, processes, and data using RAG memory.
An agent that writes proposals is different from one that processes invoices. Each is purpose-built.
Every interaction, every rating, every knowledge update compounds.
Drop any agent into any app, tool, or workflow via its dedicated API key.
Agents collaborate through workflows and the CEO Agent's orchestration.
The difference between "using AI" and having an AI that works for your business.
One form. A few decisions. An agent ready to work. Here's what you'll fill in.
Give your agent an identity. This is how you'll find and reference this agent across the platform — in workflows, in the CEO Agent's roster, and in your agent management dashboard.
Examples:
Best practice: Use descriptive, role-based names. Think of it like a job title — clear enough that anyone on your team knows what this agent does at a glance.
Tell the system (and your team) what this agent is for. A brief description of the agent's purpose, domain, and capabilities. This helps the CEO Agent make better decisions when selecting agents for project sub-tasks.
Examples:
Best practice: Be specific about what the agent does, what knowledge domain it operates in, and what kind of output it produces.
Organize your agents by function. Categories help you find agents quickly and help the CEO Agent understand the agent's domain.
Choose the AI model that powers this agent. Different models have different strengths:
| Model | Best For |
|---|---|
| Claude Sonnet 4.5 | Complex reasoning, nuanced writing, detailed analysis |
| GPT-4o | General purpose, fast, good balance of speed and quality |
| Claude Haiku | Fast, cost-efficient tasks, high-volume processing |
| Other available models | Varies by release |
Best practice: Start with a capable model like Claude Sonnet 4.5 for complex agents. For high-volume, simpler tasks, use a faster model to conserve credits. You can always clone the agent onto a different model later.
Pro tip: When a new model is released, don't update your working agent — clone it onto the new model and test. Keep the original as your reliable fallback.
Architect or Executor — define this agent's role in the system.
Designs systems, creates specifications, generates task lists, makes technical decisions.
Used by CEO Agent for: Project architecture, spec creation, system design
Completes specific tasks — writes code, generates content, processes data, handles integrations.
Used by CEO Agent for: Individual sub-tasks assigned by the CEO Agent
Best practice: Most agent libraries are heavily executor-weighted. A typical setup might be 2–4 architects and 10–20+ executors. Architects are generalists with deep knowledge; executors are specialists with focused capabilities.
Define this agent's personality, rules, and expertise. The system prompt is the instruction set that defines HOW the agent behaves — its role, tone, rules, constraints, output format, and quality standards.
Example — Sales Proposal Writer:
You are an expert sales proposal writer for [Company Name]. You write compelling, professional proposals that follow our standard format. Rules: - Always include an executive summary, scope of work, timeline, pricing, and next steps - Use confident but not aggressive language - Reference relevant case studies when available - Include specific ROI projections based on the client's situation - Never make promises about timelines without qualification Output format: Markdown with clear section headers.
Best practice: Be explicit about what you want and don't want. The more precise your system prompt, the more consistent your agent's output. Think of it as the agent's job description + performance expectations.
Define the input format — what users send to this agent. The user prompt template defines what information this agent expects when it's called. It MUST include at least one variable in the format ${variableName}.
Why variables matter: Variables are how your agents receive different inputs each time they run. When this agent is called — whether manually, via API, or through a workflow — the variable is populated with the specific data for that execution.
Sales Proposal Writer:
Write a sales proposal for the following client:
Client: ${clientName}
Industry: ${clientIndustry}
Requirements: ${clientRequirements}
Budget Range: ${budgetRange}
Customer Support Agent:
A customer has submitted the following support request.
Respond helpfully using our knowledge base.
Customer: ${customerName}
Issue: ${customerIssue}
Account Type: ${accountType}
Best practice: Include enough variables to give the agent the context it needs for quality output. Don't over-engineer — you can always add more variables by cloning and updating the agent.
Give your agent a face. Upload an avatar or image to visually identify this agent in your dashboard, workflows, and team views. Optional but recommended — especially when your team is working with 10+ agents, visual identification speeds up navigation significantly.
It appears in your Agent Manager dashboard. It's available to the CEO Agent for project assignments. It has its own API key for external access. And it's ready for RAG training to make it an expert in your specific domain.
Never modify a working agent when you can clone it first. Cloning is how power users build, test, and iterate — fast and safe.
Cloning creates an exact copy of an existing agent — same name (with "Clone" appended), same description, same prompts, same model, same type. You then modify the clone without touching the original.
Start with a battle-tested template agent (like "Sales Proposal Writer - Standard"). Clone it. Update the system prompt with your company's specific language, pricing, and case studies. Now you have a personalized agent built on a proven foundation.
Your agent runs perfectly on Claude Sonnet 4.5. A new model just released. Clone the agent, change the model to the new one, and test. If it's better, switch. If not, your original is untouched.
Your Content Writer agent is excellent. But you need one for blog posts and one for email campaigns. Clone it twice. Update one clone for blog-specific writing. Update the other for email-specific writing. Three agents, one original effort.
Clone an agent, make one change (different system prompt, different model, different user prompt structure), and run both on the same inputs. Compare outputs. Keep the winner.
About to make a significant change to an agent that's working well? Clone it first. If the changes don't work out, the original is right there, unchanged and ready to go.
The rule of thumb: If it's working, don't touch it. Clone it.
The clone inherits everything except RAG memories (which stay with the original). You can add new RAG memories to the clone independently.
RAG training is how your agents go from "smart AI assistant" to "knows our business inside and out." And it takes minutes, not months.
RAG stands for Retrieval-Augmented Generation. In plain English:
You upload your company's documents. The agent reads them, remembers them, and uses that knowledge every time it responds.
It's the difference between asking a stranger for advice and asking someone who's read every document your company has ever produced.
Three steps. Your agent now has expertise it didn't have 60 seconds ago.
# Add a single file
ceo addRag ./pricing-guide.md
# Add an entire folder (recursively)
ceo addRagDirectory ./docs/ --recursive
Processes individual files or recursively ingests entire folder structures.
| Agent Type | High-Value RAG Knowledge |
|---|---|
| Sales Agents | Pricing guides, case studies, proposal templates, competitor analysis, client FAQs |
| Support Agents | Knowledge base articles, troubleshooting guides, product docs, policy documents |
| Architect Agents | API docs, coding standards, infra preferences, past specs, tech stack docs |
| Content Agents | Brand guidelines, tone examples, past content that performed well, style guides |
| Operations Agents | Process docs, SOPs, compliance requirements, reporting templates |
Every document you upload makes the agent better — not just for the current task, but for every future task. An architect agent that learned your Salesforce field mappings on Project 1 remembers them on Project 5. A support agent that learned your refund policy on Day 1 applies it correctly on Day 100.
Your agents accumulate institutional knowledge. They don't quit. They don't forget. They don't need to be retrained when you hire a new team member.
Full Agent Roster
Name, type, category, model, status
Quick Access
Edit config, prompts, or add RAG memories
Clone Instantly
One-click clone from any agent
API Key Access
View and copy dedicated API keys
Performance Visibility
Ratings, task history, CEO Agent projects
Search & Filter
By name, category, type, or model
Think of it as your AI org chart. Every agent has a role, capabilities, and a track record — all visible from one screen.
Embed in any app — call from any frontend, mobile app, or web service
Connect to any workflow — trigger agents from external systems via HTTP
Build products — create AI-powered services using trained agents as the intelligence layer
1-to-1 mapping — each agent has its own key for granular access and usage management
Quick Example
const { CeoAI } = require('@ceo-ai/sdk');
const ceo = new CeoAI({ apiKey: 'sk_live_your_api_key_here' });
// Kick off an agent with structured client context
const { response, metadata } = await ceo.promptAndWait(
'Build a proposal for Acme Corp (Manufacturing): ' +
'Automated inventory management with AI-powered reorder predictions'
);
console.log(response);
// => { answer: "Here's a tailored proposal for Acme Corp's inventory system..." }For agencies and developers: This is how you turn CEO.ai into a backend for AI-powered products. Train an agent on a client's domain, give it RAG knowledge, and embed it via API into whatever they need — a chatbot, a processing pipeline, a customer-facing app. You focus on the frontend and the client relationship. CEO.ai handles the AI.
Pre-built agent templates for common roles and functions. Clone one, update the prompts, add your RAG knowledge, and you have a custom agent in minutes — not hours.
| Template | Type | What It Does |
|---|---|---|
| General Architect | Architect | Designs full-stack application architectures from project descriptions |
| Content Writer | Executor | Writes marketing content, blog posts, and copy based on brand guidelines |
| Data Transformer | Executor | Takes unstructured data and transforms it into structured formats |
| API Integration Specialist | Executor | Builds connections between platforms using REST APIs |
| Code Reviewer | Executor | Reviews code for quality, security, and best practices |
| Documentation Writer | Executor | Generates technical and user documentation from code and specs |
| Template | Type | What It Does |
|---|---|---|
| Sales Proposal Writer | Executor | Generates customized proposals based on client requirements and your pricing |
| Customer Support Tier 1 | Executor | Handles first-touch support using your knowledge base |
| Operations Analyst | Executor | Analyzes operational data and generates insights and recommendations |
| Meeting Summarizer | Executor | Processes transcripts into structured summaries with action items |
| HR Onboarding Assistant | Executor | Guides new hires through onboarding using your company's materials |
| Financial Report Generator | Executor | Generates formatted financial reports from raw data |
| Workflow Architect | Architect | Designs multi-step operational workflows with agent assignments |
| Integration Architect | Architect | Designs multi-platform integration architectures |
Templates are starting points, not finished products. The magic happens when you combine a proven template structure with your company's specific knowledge via RAG training.
Everything you create is private. No other customer can see or access your agents.
You can request to whitelist specific agents to the Community Agents marketplace.
Once approved, your agent becomes available for other customers' CEO Agent projects.
When your agent is selected for another customer's task, you earn credits.
If your agents contain proprietary logic in their prompts or represent competitive advantages, keep them private. That's the default, and there's zero pressure to change it.
Every plan includes guided setup — we don't just give you a form and wish you luck. On your setup call, we'll identify your highest-value agent roles, help you write the prompts, and get RAG training started.
By the end of your first week, you'll have a roster of custom AI agents that know your business and are ready to work.
Most customers have 3–5 custom agents live within their first week.