Agent Configuration
Model selection, temperature, tools, knowledge base setup
Agent Configuration
Configure your AI agents to perform exactly as needed by selecting the right model, adjusting behavior parameters, enabling tools, and providing context through knowledge bases. Proper configuration ensures your agents work effectively and consistently.
Model Selection
Choose the AI model that powers your agent based on your needs and budget.
Available Models
GPT-4 Turbo
- Best for: Complex reasoning, detailed analysis, and multi-step tasks
- Strengths: Excellent instruction following, broad knowledge
- Speed: Moderate (2-5 seconds per response)
- Cost: Higher tier
GPT-3.5 Turbo
- Best for: Quick responses, simple categorization, routine tasks
- Strengths: Fast, cost-effective, reliable
- Speed: Fast (1-2 seconds per response)
- Cost: Lower tier
Claude 3 Opus
- Best for: Long-context analysis, detailed writing, complex workflows
- Strengths: Excellent reasoning, strong safety guardrails
- Speed: Moderate (2-4 seconds per response)
- Cost: Higher tier
Claude 3 Sonnet
- Best for: Balanced performance and cost
- Strengths: Good reasoning, fast responses, cost-effective
- Speed: Fast (1-3 seconds per response)
- Cost: Mid tier
Claude 3 Haiku
- Best for: High-volume, simple tasks
- Strengths: Very fast, extremely cost-effective
- Speed: Very fast (under 1 second per response)
- Cost: Lower tier
Changing the Model
- Navigate to your agent settings
- Click Model Configuration
- Select your preferred model from the dropdown
- Click Save Changes
The agent will use the new model for all future interactions. Existing task history remains unchanged.
** Pro Tip**: Start with a mid-tier model like Claude 3 Sonnet or GPT-3.5 Turbo. Upgrade to premium models if you need better reasoning or analysis capabilities.
Temperature Settings
Temperature controls how creative or deterministic your agent's responses are.
Understanding Temperature
Temperature: 0.0 - 0.3 (Focused)
- Responses are consistent and predictable
- Best for: Data extraction, categorization, structured output
- Use when: You need the same input to produce nearly identical output
Temperature: 0.4 - 0.7 (Balanced)
- Mix of consistency and creativity
- Best for: General task management, email processing, summary generation
- Use when: You want reliable results with some variation
Temperature: 0.8 - 1.0 (Creative)
- More varied and creative responses
- Best for: Content generation, brainstorming, exploratory analysis
- Use when: You want diverse ideas and creative solutions
Temperature: >1.0 (Experimental)
- Highly unpredictable and creative
- Best for: Creative writing, unconventional ideas
- Use when: Maximum variety is more important than consistency
Setting Temperature
- Open agent settings
- Navigate to Behavior Settings
- Adjust the Temperature slider
- Test with sample inputs
- Save when satisfied with results
** Important**: High temperature (>0.8) may produce inconsistent results for structured tasks like data extraction or categorization.
Tool Configuration
Tools define what actions your agent can perform in Verk.
Available Tools
Task Management
create_task- Create new tasksupdate_task- Modify existing tasksdelete_task- Remove taskssearch_tasks- Find tasks by criteria
Project Management
create_project- Create new projectsupdate_project- Modify project detailslist_projects- View available projects
Assignment & Organization
assign_task- Assign tasks to team membersadd_labels- Add tags to tasksset_priority- Set task priority levelsset_due_date- Schedule task deadlines
Information Retrieval
get_task_details- Retrieve full task informationsearch_knowledge_base- Query uploaded documentslist_team_members- See available assignees
Enabling Tools
- Go to agent settings → Tools
- Review available tools
- Toggle on the tools your agent needs
- Click Save Configuration
Example configuration for "Email Processing Agent":
create_task- To create tasks from emailsassign_task- To assign to team membersadd_labels- To categorize by typeset_priority- To mark urgent itemsdelete_task- Email agent shouldn't delete taskscreate_project- Email agent doesn't create projects
** Best Practice**: Only enable tools your agent actually needs. Fewer tools reduce confusion and improve performance.
Tool Permissions
Control which projects and tasks your agent can modify:
All Projects
- Agent can access any project in the organization
- Best for: General-purpose agents
Specific Projects
- Limit agent to selected projects only
- Best for: Department-specific or client-specific agents
Configuration:
- Agent settings → Permissions
- Select All Projects or Specific Projects
- If specific, choose projects from the list
- Save changes
Knowledge Base Setup
Upload documents to give your agent context and reference information.
Adding Documents
- Navigate to agent settings → Knowledge Base
- Click Upload Documents
- Select files or drag and drop
- Wait for processing to complete
- Verify documents appear in the knowledge base list
Supported formats:
- PDF documents
- Microsoft Word (.docx)
- Plain text (.txt)
- Markdown (.md)
- CSV spreadsheets
- Web pages (paste URL)
Document Types
Reference Materials
- Company policies and procedures
- Product documentation
- Style guides and templates
- FAQs and knowledge articles
Contextual Information
- Client requirements and briefs
- Project specifications
- Meeting notes and decisions
- Historical data and past examples
Structured Data
- Team member expertise and availability
- Project templates and structures
- Priority frameworks
- Naming conventions
Knowledge Base Limits
Free Plan:
- 10 documents per agent
- 5 MB max file size
- 50 pages total across all documents
Pro Plan:
- 100 documents per agent
- 20 MB max file size
- 500 pages total per agent
Enterprise Plan:
- Unlimited documents
- 100 MB max file size
- Unlimited pages
Managing Knowledge Base
Update documents:
- Navigate to Knowledge Base
- Click the document to update
- Click Replace to upload a new version
- The agent will use the latest version
Remove documents:
- Find the document in the list
- Click the three-dot menu
- Select Remove from Knowledge Base
- Confirm deletion
Search knowledge base: Test what information your agent can find:
- Click Test Knowledge Base
- Enter a sample query
- Review retrieved information
- Adjust documents if needed
** Pro Tip**: Keep documents focused and well-organized. Multiple small, topic-specific documents work better than one large document.
Advanced Configuration
Response Length
Control how verbose your agent's responses are:
Concise (50-100 words)
- Brief summaries and quick answers
- Best for: Status updates, quick categorization
Standard (100-300 words)
- Balanced detail and brevity
- Best for: Most use cases
Detailed (300-500 words)
- Comprehensive explanations
- Best for: Complex analysis, documentation
Very Detailed (500+ words)
- Extensive analysis and documentation
- Best for: Research, comprehensive reports
Custom Instructions Priority
When an agent has both system-level and custom instructions:
- Custom instructions take priority
- System instructions provide baseline behavior
- Tool configurations override both if there are conflicts
Rate Limiting
Prevent agents from overwhelming your system:
Actions per hour:
- Free: 50 actions/hour
- Pro: 500 actions/hour
- Enterprise: Unlimited
Actions per day:
- Free: 200 actions/day
- Pro: 2,000 actions/day
- Enterprise: Unlimited
Concurrent operations:
- All plans: Maximum 5 simultaneous operations
Retry Configuration
Configure how agents handle failures:
Automatic Retries:
- Default: 3 retry attempts
- Retry delay: 2 seconds (exponential backoff)
- Max retry delay: 30 seconds
Custom retry settings:
- Agent settings → Advanced → Retry Logic
- Set maximum retry attempts (0-5)
- Configure retry delay (1-60 seconds)
- Enable/disable exponential backoff
Configuration Templates
Save and reuse agent configurations:
Saving Templates
- Configure an agent completely
- Click Save as Template
- Enter template name and description
- Choose sharing (private or organization-wide)
- Click Create Template
Using Templates
- Create a new agent
- Click Start from Template
- Select a template
- Customize as needed
- Save the agent
Popular template configurations:
Email Processor
- Model: GPT-3.5 Turbo
- Temperature: 0.5
- Tools: create_task, assign_task, add_labels, set_priority
- Knowledge base: Team assignments, priority guidelines
Meeting Notes Analyzer
- Model: GPT-4 Turbo
- Temperature: 0.3
- Tools: create_task, assign_task, set_due_date
- Knowledge base: Project contexts, team member roles
Content Categorizer
- Model: Claude 3 Haiku
- Temperature: 0.2
- Tools: add_labels, update_task
- Knowledge base: Category definitions, tagging guidelines
Research Assistant
- Model: Claude 3 Opus
- Temperature: 0.7
- Tools: create_task, search_tasks, search_knowledge_base
- Knowledge base: Research guidelines, citation formats
Testing Configuration
Verify your agent works as expected before deploying:
Test Mode
- Navigate to agent settings → Testing
- Enable Test Mode
- Enter sample inputs
- Review agent responses and actions
- Adjust configuration based on results
Test scenarios to try:
Task Creation:
- Input: Typical email or message
- Expected: Task created with correct fields
Categorization:
- Input: Various task descriptions
- Expected: Appropriate labels applied
Assignment:
- Input: Task requiring specific expertise
- Expected: Assigned to correct team member
Priority Setting:
- Input: Tasks with different urgency levels
- Expected: Correct priority assigned
Test Logs
Review detailed logs of agent behavior:
- Agent settings → Test Logs
- View input, reasoning, and actions taken
- Check for errors or unexpected behavior
- Export logs for detailed analysis
Troubleshooting Configuration
Agent Not Responding
Check:
- Model is selected and available
- Agent is enabled (not paused)
- Rate limits not exceeded
- Tools are properly enabled
Fix:
- Review agent status in settings
- Check recent activity logs
- Verify API credentials if using external models
- Contact support if issues persist
Inconsistent Results
Common causes:
- Temperature too high (>0.8)
- Instructions too vague
- Conflicting instructions
- Insufficient knowledge base context
Fix:
- Lower temperature to 0.3-0.5
- Make instructions more specific
- Review instructions for conflicts
- Add relevant documents to knowledge base
Wrong Actions Performed
Common causes:
- Too many tools enabled
- Unclear instructions about when to use tools
- Missing permission restrictions
Fix:
- Disable unnecessary tools
- Add specific guidelines about tool usage
- Restrict agent to specific projects
- Add approval workflows for critical actions
Slow Performance
Common causes:
- Large knowledge base (>100 documents)
- High-complexity model (GPT-4, Claude Opus)
- Too many concurrent operations
Fix:
- Reduce knowledge base size
- Switch to faster model for simple tasks
- Implement rate limiting
- Split complex agents into specialized agents
Best Practices
Start Simple Begin with minimal configuration and add complexity as needed:
- Choose a mid-tier model
- Enable only essential tools
- Add 2-3 key documents to knowledge base
- Test thoroughly before expanding
Monitor and Iterate Regularly review agent performance:
- Check activity logs weekly
- Review success rates for key actions
- Gather feedback from team members
- Adjust configuration based on results
Document Configuration Keep records of why you configured agents certain ways:
- Add notes in agent description
- Document template purposes
- Track configuration changes
- Share learnings with team
Security Considerations Protect sensitive information:
- Limit tool access to necessary projects
- Review knowledge base for sensitive data
- Use test mode before production deployment
- Monitor for unexpected behavior
Cost Optimization Balance performance and budget:
- Use faster, cheaper models for simple tasks
- Reserve premium models for complex reasoning
- Monitor usage and adjust rate limits
- Review and remove unused agents
Related Documentation
- Creating AI Agents - Set up your first agent
- Knowledge Base Management - Detailed guide to documents
- Agent Analytics - Monitor agent performance
- Automation Rules - Combine agents with workflows
Need help configuring an agent? Contact our support team or check the FAQ.