Using Personas

Execute structured AI tasks with built-in approval workflows.

What We're Building

A persona-based workflow that:

• Gathers structured inputs
• Follows a defined SOP
• Produces specific deliverables
• Requires approval at checkpoints

What are Personas?

Personas are pre-configured AI specialists designed for specific tasks. Unlike open-ended chat agents, personas:

Feature	Agents	Personas
Input	Freeform chat	Structured fields
Output	Conversation	Defined deliverables
Process	Dynamic	SOP-based
Approval	Optional	Built-in checkpoints

Step 1: Browse Available Personas

1. Navigate to Personas in the sidebar
2. Browse by category:
   • Research — Market research, competitor analysis
   • Content — Blog posts, social media, documentation
   • Development — Code review, architecture, testing
   • Data — Analysis, reporting, visualization
   • Business — Strategy, proposals, planning

Step 2: Select a Persona

For this guide, we'll use Blog Post Writer:

{
  name: "Blog Post Writer",
  category: "content",
  description: "Creates engaging blog posts from topic briefs",
  model: "claude-sonnet-4-5"
}

Step 3: Start a New Task

1. Click Run Task
2. Fill in the structured inputs:

Field	Type	Example
Topic	text	"Introduction to RAG"
Target Audience	select	"Technical developers"
Tone	select	"Professional but approachable"
Word Count	number	1500
Key Points	multiline	"What RAG is, How it works, Use cases"
Reference URLs	url_list	"https://docs.example.com/rag"

Step 4: Monitor Progress

The persona follows its SOP:

✓ Step 1: Research topic and references
✓ Step 2: Create outline
→ Step 3: Write first draft (in progress)
○ Step 4: Review and revise
○ Step 5: Final polish

Each step shows:

• Status (pending, in progress, complete)
• Time spent
• Output preview

Step 5: Review at Checkpoints

At approval checkpoints, the persona pauses:

Checkpoint: Outline Review

The outline is ready for your review:

1. Introduction - What is RAG?
2. The Problem with Traditional LLMs
3. How RAG Works
   3.1 Retrieval
   3.2 Augmentation
   3.3 Generation
4. Implementation Examples
5. Best Practices
6. Conclusion

[Approve] [Request Changes] [Cancel]

Options:

• Approve — Continue to next step
• Request Changes — Provide feedback, persona revises
• Cancel — Stop the task

Step 6: Receive Deliverables

When complete, access your outputs:

{
    deliverables: [
        {
            type: "document",
            name: "blog-post-rag-intro.md",
            format: "markdown"
        },
        {
            type: "document",
            name: "social-snippets.txt",
            format: "text"
        },
        {
            type: "summary",
            name: "seo-keywords.json",
            format: "json"
        }
    ];
}

Download or copy each deliverable as needed.

Approval Workflows

Approval Levels

Level	Behavior
none	Runs autonomously to completion
checkpoints	Pauses at defined review points
all_steps	Requires approval after each step

Setting Approval Level

When starting a task:

{
  approvalLevel: "checkpoints",
  notifyOn: ["checkpoint", "completion"],
  notifyChannels: ["email", "slack"]
}

Responding to Approvals

Approvals can be handled via:

1. Dashboard — Click notification, review, approve
2. Email — Reply with "Approved" or feedback
3. Slack — Use button actions in message
4. API — Programmatically approve

POST /api/personas/tasks/{taskId}/approve
{
  "decision": "approved",
  "feedback": "Looks good, proceed."
}

Cost Tracking

Monitor resource usage:

{
  task: "Blog Post Writer - RAG Introduction",
  cost: {
    tokens: 45000,
    credits: 12.5,
    estimatedUsd: 0.45
  },
  duration: "18 minutes"
}

Set cost limits when starting tasks:

{
  costLimits: {
    maxCredits: 50,
    maxDuration: "1 hour",
    pauseOnLimit: true  // Pause instead of fail
  }
}

Example: Research Report

Using the Market Research persona:

Inputs:

{
  company: "Acme Corp",
  industry: "SaaS",
  researchAreas: ["Market size", "Competitors", "Trends"],
  depth: "comprehensive",
  sources: ["public_data", "news", "reports"]
}

SOP Steps:

1. Gather company background
2. Analyze market landscape
3. Identify key competitors
4. Research industry trends
5. Synthesize findings
6. Generate report

Checkpoints:

• After competitor analysis (verify competitors)
• Before final report (review draft)

Deliverables:

• Market research report (PDF)
• Competitor comparison table (spreadsheet)
• Key insights summary (document)

Creating Custom Personas

Build your own personas:

1. Go to Personas > Create Persona
2. Define inputs, SOP, and deliverables
3. Configure approval checkpoints
4. Test with sample tasks
5. Share with your team

See Creating Personas for details.

Best Practices

Input Design

• Use specific field types (select vs freeform)
• Provide helpful descriptions
• Set sensible defaults
• Mark required vs optional fields

SOP Design

• Break work into clear steps
• Add checkpoints at decision points
• Include time estimates
• Handle edge cases

Approval Strategy

• More checkpoints = more control, slower
• Fewer checkpoints = faster, less oversight
• Match approval level to task criticality

Troubleshooting

Issue	Solution
Task stuck at checkpoint	Check notifications, approve or cancel
Cost exceeded	Increase limit or optimize persona
Poor quality output	Refine inputs, add examples
Persona not finding info	Add reference URLs or documents

Next Steps

• Creating Personas — Build custom personas
• Persona reference — All available personas
• Using Personas — Complete persona workflow guide