Creating an Assistant
Build custom assistants tailored to your specific needs:1
Access Assistant Creator
- Click ”+” button in Assistants section
- Or use “Create Assistant” from space settings
- Choose “Custom Assistant” option
2
Basic Information
- Name: Give your assistant a descriptive name
- Description: Explain its purpose and capabilities
- Logo: Upload or select an icon
- Category: Classify for easy discovery
3
Configure Instructions
Write system prompt defining behavior, persona, and capabilities
4
Select Tools
Enable tools the assistant can use
5
Model Settings
Configure AI model and parameters
6
Test & Save
Test with sample prompts and save
Writing Instructions
The instructions field is the core of your assistant’s behavior:Structure
Best Practices
- Be specific and detailed
- Include concrete examples
- Define clear boundaries
- Specify tone and style
- Test with edge cases
- Iterate based on results
Example Instructions
Customer Support Assistant
Customer Support Assistant
Content Writer Assistant
Content Writer Assistant
Tool Configuration
Enable and configure tools for your assistant:Available Tools
Web Search
Search the internet for current information
Space Search
Search within space data and documents
URL Similarity
Find similar content from URLs
API Request
Make HTTP requests to APIs
Model Selector
Choose models dynamically
Add Data
Access and query custom data
Image Generation
Create images with DALL-E
MCP Tools
Custom tool integrations
Tool Configuration
Each tool can be configured:- Enable/Disable: Turn tools on or off
- Permissions: Control tool access
- Parameters: Set tool-specific settings
- Usage Limits: Restrict frequency or scope
Tool Selection Guidelines
Enable tools based on assistant needs:- Customer Support: Web Search, Space Search
- Content Writer: Web Search, URL Similarity
- Data Analyst: Space Search, Add Data, API Request
- Research: Web Search, URL Similarity, Space Search
- Developer: Web Search, API Request, Model Selector
Model Configuration
Configure AI model settings:Model Selection
Choose the primary model:- GPT-4: Most capable, higher cost
- GPT-3.5: Fast and cost-effective
- Claude: Excellent for analysis
- Gemini: Google’s latest models
- Smart Model: Automatic selection
Parameter Settings
Temperature (0-2)
Temperature (0-2)
Controls randomness and creativity:
- 0-0.3: Focused, consistent, deterministic
- 0.4-0.7: Balanced creativity and consistency
- 0.8-1.5: More creative and varied
- 1.6-2.0: Highly creative, less predictable
- Customer Support: 0.3
- Creative Writing: 1.0
- Code Generation: 0.2
- General Use: 0.7
Max Tokens
Max Tokens
Maximum response length:
- Controls how long responses can be
- Higher = longer responses, more cost
- Set based on expected response needs
- Short answers: 500-1000
- Detailed responses: 2000-4000
- Long-form content: 4000-8000
Top P (0-1)
Top P (0-1)
Nucleus sampling threshold:
- Controls diversity of word choices
- 1.0 = consider all possibilities
- Lower = more focused selection
Frequency Penalty (0-2)
Frequency Penalty (0-2)
Reduces repetition:
- 0 = no penalty
- Higher = less repetition
- Too high = unnatural text
Presence Penalty (0-2)
Presence Penalty (0-2)
Encourages topic diversity:
- 0 = no penalty
- Higher = more topic variation
- Use for creative tasks
Assistant Metadata
Configure organizational details:Categories
Choose appropriate category:- Customer Support
- Content Creation
- Data Analysis
- Code Development
- Research
- Sales & Marketing
- Personal Productivity
- HR & Recruitment
- Custom
Visibility
Set who can discover your assistant:- Private: Only you
- Space: Space members
- Organization: All organization members
- Public: Discoverable by anyone (requires approval)
Tags
Add searchable keywords:- Help with discovery
- Group related assistants
- Indicate capabilities
- Specify industries or use cases
Testing Your Assistant
Validate assistant behavior:Test Prompts
Create test cases covering:- Happy Path: Standard use cases
- Edge Cases: Unusual inputs
- Error Handling: Invalid requests
- Tool Usage: Verify tool calls
- Tone & Style: Check communication
Iteration Process
- Test with diverse prompts
- Identify issues or improvements
- Adjust instructions or settings
- Re-test
- Collect user feedback
- Continue iterating
Version Management
Track assistant changes:Creating Versions
- Save major changes as new versions
- Add version notes
- Tag stable releases
- Keep history for rollback
Version Naming
Use clear naming conventions:- Semantic versioning (1.0, 1.1, 2.0)
- Date-based (2024-01-15)
- Descriptive names (initial, improved, stable)
Sharing Your Assistant
Make assistant available to others:Share with Spaces
- Open assistant configuration
- Click “Share with Spaces”
- Select target spaces
- Set permissions
- Save
Share with Users
- Go to sharing settings
- Enter user emails
- Assign permissions
- Send invitations
Publishing Organization-Wide
- Configure assistant
- Submit for organization publishing
- Wait for admin approval
- Assistant becomes available to all
Advanced Configuration
Human Intervention
Configure when to involve humans:- Escalation triggers
- Approval workflows
- Review requirements
- Notification settings
Knowledge Cutoff
Specify when assistant’s knowledge ends:- Important for time-sensitive information
- Helps users understand limitations
- Guides when to use web search
Output Formatting
Define response format:- Markdown for rich text
- JSON for structured data
- Plain text for simplicity
- HTML for web content