Common Workflows
Step-by-step guides for common administrative tasks in Outermind.
Overview
This guide provides detailed workflows for common tasks you'll perform as an Outermind administrator. Each workflow includes prerequisites, step-by-step instructions, and verification steps.
Agent Management Workflows
Creating a New Agent
Goal: Set up a new specialist agent for email processing
Prerequisites:
- Active LLM provider
- Understanding of agent's purpose
- List of tools agent will need
Steps:
- Navigate to Configuration > AI Agents > Agents
- Click Add Agent
- Fill in basic information:
- Name: Descriptive name (e.g., "Sales Inquiry Agent")
- Description: Brief purpose summary
- Select LLM model based on task complexity
- Set temperature (0.7 recommended for most cases)
- Write instructions following the template in Agents documentation
- Click Improve with AI for suggestions
- Select required tools
- Click Save
Verification:
- Agent appears in list with "Active" status
- Test by sending email to assigned mailbox
- Review execution for expected behavior
Cloning an Existing Agent
Goal: Create a new agent based on an existing one
Steps:
- Go to Configuration > AI Agents > Agents
- Find the agent to clone
- Click the clone icon (or open and copy settings)
- Modify name and description
- Adjust instructions as needed
- Update tool assignments if different
- Save the new agent
When to clone:
- Creating similar agents for different teams
- Testing instruction changes safely
- Setting up development/production pairs
Updating Agent Instructions
Goal: Improve agent behavior through instruction changes
Prerequisites:
- Identified issue or improvement area
- Example executions showing current behavior
Steps:
- Go to Configuration > AI Agents > Agents
- Click on the agent to edit
- Review current instructions
- Make targeted changes:
- Add specific guidance for problem scenarios
- Clarify ambiguous instructions
- Add examples if helpful
- Click Improve with AI for suggestions (optional)
- Save changes
- Test with a representative email
Best practices:
- Make one change at a time
- Document what you changed and why
- Test before and after with same input
- Monitor for unintended effects
Mailbox Management Workflows
Adding a New Monitored Mailbox
Goal: Start monitoring a new email address
Prerequisites:
- Microsoft 365 shared mailbox created
- Active agent to assign
- Appropriate Azure AD permissions
Steps:
- Navigate to Configuration > Email Processing > Monitored Mailboxes
- Click Add Mailbox
- Enter mailbox email address
- Set display name
- Select the agent to process emails
- Configure options:
- Auto-reply enabled/disabled
- Include thread context (recommended)
- Save the mailbox
- Click Manage Subscriptions
- Click Create Subscription
- Verify subscription shows "Active"
Verification:
- Send test email to the mailbox
- Check executions within 1-2 minutes
- Verify response if auto-reply enabled
Changing an Agent Assignment
Goal: Switch which agent processes a mailbox
Steps:
- Go to Configuration > Email Processing > Monitored Mailboxes
- Click on the mailbox
- Change the assigned agent
- Save changes
Note: Changes take effect immediately for new emails. In-progress executions use the previous agent.
Pausing Email Processing
Goal: Temporarily stop processing emails
For one mailbox:
- Go to the mailbox settings
- Set status to inactive
- Save
For all mailboxes:
- Go to Account Settings
- Enable "Pause Agent Execution"
- Save
Note: Emails received while paused will not be processed. They won't be automatically picked up when resumed.
Troubleshooting Workflows
Investigating a Failed Execution
Goal: Understand why an execution failed
Steps:
- Go to Operations > Activity > Agent Executions
- Filter by status: "Failed"
- Click on the failed execution
- Review the timeline:
- What was the input?
- Where did it fail?
- What error message appeared?
- Click Troubleshoot with AI for analysis
- Identify the root cause
- Take corrective action
Common causes:
- LLM provider unavailable
- Tool configuration error
- Invalid response format
- Timeout
Finding Expensive Executions
Goal: Identify high-cost executions for optimization
Steps:
- Go to Operations > Activity > Agent Executions
- Sort by cost (descending)
- Review top executions:
- Were they legitimate complex tasks?
- Are there patterns (same agent, same type)?
- For high-cost executions:
- Check token counts
- Review tool calls (many calls = higher cost)
- Check response length
- Identify optimization opportunities:
- Shorter instructions
- Smaller model for simple tasks
- Fewer unnecessary tool calls
Comparing Agent Performance
Goal: Determine which agent version performs better
Steps:
- Create a clone of the agent
- Make the changes to the clone
- Set up A/B test:
- Route some traffic to original
- Route some to clone
- Collect execution data over time
- Compare metrics:
- Success rate
- Cost per execution
- Response quality (manual review)
- Choose the better performer
- Deactivate or delete the other
VIP Management Workflows
Adding VIP Contacts
Goal: Ensure important contacts receive priority handling
Steps:
- Go to Configuration > Contacts & Directory > VIPs
- Click Add VIP
- Enter contact information:
- Email address (required)
- Name
- Company
- Set privilege level (1-5, higher = more important)
- Add notes about special handling
- Save
Usage in agents: Add to agent instructions:
Before processing, check if the sender is a VIP using lookup_vip.
If they are VIP level 3 or higher, prioritize their request and
ensure a thorough, personalized response.
Bulk Importing VIPs
Goal: Add multiple VIP contacts at once
Steps:
- Prepare a CSV file with columns:
- email, name, company, privilege_level, notes
- Go to Configuration > Contacts & Directory > VIPs
- Click Import
- Upload CSV file
- Map columns to fields
- Review preview
- Import
Knowledge Management Workflows
Adding a Search Index
Goal: Make new content searchable by agents
Steps:
- Prepare your content (documents, web pages, etc.)
- Go to Configuration > Knowledge Base > Knowledge Sources
- Add the source:
- SharePoint: Connect to document library
- Web: Provide URLs to crawl
- Upload: Upload files directly
- Configure crawl settings
- Start initial crawl
- Go to Search Indexes
- Create or update index with new source
- Verify by testing searches
Re-indexing Content
Goal: Update search index with new or changed content
Steps:
- Go to Configuration > Knowledge Base > Index Health > Scan Monitoring
- Find the relevant source
- Click Rescan or Full Index
- Monitor progress
- Verify changes in search results
System Administration Workflows
Reviewing System Health
Goal: Ensure all components are working properly
Daily check:
- View Operations > Dashboard > Home
- All components should show connected
- Check LLM Providers
- All should show green status
- Check Monitored Mailboxes
- Subscriptions should be active
- Review Operations > Activity > Agent Executions
- Check for unusual failure rates
- Check Usage
- Verify costs are expected
Setting Up a New Administrator
Goal: Grant admin access to a new user
Steps:
- Ensure user has Azure AD account
- Go to Admin > Access Control > Users
- Click Add User
- Select the Azure AD user
- Assign the "Admin" role
- Save
User will need to:
- Sign in with their Azure AD credentials
- Accept any consent prompts
- Access the Admin Console URL
Backing Up Configuration
Goal: Preserve system configuration for recovery
Recommended backup items:
- Agent configurations (instructions, tool assignments)
- VIP lists
- Search index configurations
- Provider settings (excluding API keys)
Manual backup:
- Export agent instructions to documents
- Export VIP list to CSV
- Document search index settings
- Store in version control or SharePoint
Workflow Checklists
New Deployment Checklist
- LLM Provider configured and tested
- At least one agent created
- Agent instructions reviewed and tested
- Monitored mailbox added
- Subscription active
- Test email processed successfully
- VIP contacts imported (if applicable)
- Knowledge sources connected (if applicable)
- User access configured
- Monitoring in place
Weekly Maintenance Checklist
- Review failed executions
- Check provider health
- Review cost trends
- Update knowledge indexes if needed
- Address any user feedback
- Update agent instructions if needed
Monthly Review Checklist
- Full cost analysis
- Agent performance review
- Search index effectiveness
- User access audit
- Provider pricing verification
- Documentation updates