While MultiTask projects orchestrate teams of agents to build complete applications, sometimes you need a direct conversation with a single specialized agent. Whether you're prototyping an idea, getting expert advice, refining a specific component, or extending work from a project task, agent conversations give you flexible, focused AI assistance tailored to your exact needs.
When to Use Agent Conversations
⚡ Quick Tasks & Prototypes
- Generate a single component or function
- Get code review on a specific file
- Ask for architectural advice
- Explore implementation approaches
🔧 Extending Project Outputs
- Refine a task output from a MultiTask project
- Add features to generated code
- Fix issues in delivered work
- Optimize existing implementations
📚 Learning & Exploration
- Understand complex concepts
- Compare different approaches
- Get best practice recommendations
- Explore framework capabilities
🎯 Specialized Consulting
- Security audit advice
- Performance optimization suggestions
- Accessibility guidance
- Compliance requirements
Starting a Conversation
Step 1: Select Your Agent
You have two ways to access an agent for conversation:
A From the Agents Page
- 1. Navigate to the Agents page
- 2. Browse or filter to find your desired agent
- 3. Click on the agent's card to open the conversation interface
💡 Tip: Use filters to quickly find the right specialist
- • Text search for technology (e.g., "react", "postgresql")
- • Category filter for task type (e.g., "Testing/QA")
- • Model filter for specific AI capabilities
B Direct Navigation
If you know the agent's ID, go directly to:
app.ceo.ai/agents/:agentidReplace :agentid with the actual agent ID.
When you'd use this:
- • Bookmarked favorite agents
- • Shared agent links from colleagues
- • Returning to an agent you've used before
- • Integration with external tools
Step 2: Review Agent Details
On the agent conversation page, you'll see comprehensive information about your selected agent:
Agent Information
Name
The agent's title
Description
What the agent specializes in
Category
Domain or specialty area
Type
Architect or Executor
Model
AI model powering the agent
Creator
Who built this agent
Rating
Average user rating (if available)
Use this information to verify:
- You've selected the right agent for your task
- The agent's specialty matches your needs
- The agent's approach aligns with your expectations
Step 3: Configure RAG Mode
Near the top of the conversation interface, you'll see a RAG Mode toggle button.
Understanding RAG Mode
RAG (Retrieval-Augmented Generation) allows the agent to access custom memories that have been added through upskilling.
RAG Mode OFF
(Default)
Agent uses only base AI model knowledge—no access to custom memories.
RAG Mode ON
Recommended for specialized work
Agent retrieves relevant memories before responding—uses upskilled knowledge and guidelines.
When to Enable RAG Mode
Turn RAG ON when:
- • Working with proprietary systems or APIs
- • Need agent to follow specific coding standards
- • Require domain-specific expertise (compliance, protocols, etc.)
- • Building on previous upskilling you've invested in
- • Want consistent patterns across outputs
Keep RAG OFF when:
- • General questions or brainstorming
- • Exploring options without constraints
- • Quick, simple tasks
- • Base model knowledge is sufficient
- • Optimizing for speed and cost
Toggle the switch to your preferred setting before starting the conversation.
Step 4: Prepare Your Initial Prompt
You'll see an input box for entering variable text. This is where you complete the agent's user prompt template.
Understanding the User Prompt Template
When the agent was created, its creator defined a user prompt template with a variable placeholder like ${taskDescription} or ${userRequest}.
Example template (you don't see this directly):
Complete the following React component implementation:
${componentRequirements}
Requirements:
- Use TypeScript
- Include proper props interface
- Add meaningful comments
What you provide: The text that fills in the ${componentRequirements} variable.
Crafting Your Input
❌ Too vague:
"Make a button"
✅ Better:
"Create a reusable Button component with props for variant (primary, secondary, danger), size (small, medium, large), disabled state, and onClick handler. Include hover and focus states."
Conversation Best Practices
Start Strong
Your first prompt sets the tone. Provide clear, detailed requirements, include constraints and preferences, specify output format if important, and give relevant context.
Good first prompt:
"Create a TypeScript function that fetches user data from our API at /api/users/:id, handles errors gracefully with try-catch, includes retry logic for network failures (max 3 retries with exponential backoff), and returns a typed response using our UserData interface. Include JSDoc comments."
Troubleshooting
Response Taking Too Long
If waiting more than 2-3 minutes:
- • Check your internet connection
- • Verify the platform status
- • Try refreshing (may lose progress—use cautiously)
- • Contact support if persistent
Output Not What You Expected
Common causes:
- • Prompt was too vague—be more specific
- • RAG mode needed but not enabled
- • Agent specialty doesn't match task
- • Context from earlier in conversation is confusing
Solutions:
- • Clarify in follow-up
- • Restart conversation with better initial prompt
- • Try different agent
- • Toggle RAG mode
Frequently Asked Questions
Can I have multiple conversations with the same agent?
Yes! Each conversation is independent. Start as many as you need.
Are conversations private?
Yes, your conversations are private to you.
What if I disagree with the agent's output?
Provide feedback in a follow-up: "That approach won't work because [reason]. Try this instead: [alternative]"
How do I know which agent to choose?
Use filters to find your specialists. When in doubt, try a conversation and switch agents if needed.
Related Documentation
Finding Your Agents
Filter and select the right agent
Agent Upskilling
Add memories to improve RAG mode conversations
Creating Custom Agents
Build agents optimized for conversations
MultiTask Projects
When to use orchestrated projects vs. conversations
Credits & Billing
Understanding conversation credit costs
Ready to Start Your First Conversation?
Head to the Agents page, select a specialist that matches your need, and start building, learning, or solving problems with focused AI assistance.
Browse Available Agents