| title | Prompt-Based Apps |
|---|---|
| icon | terminal |
| description | Customize Omi's behavior with prompts - no server required. Create specialized AI personalities and memory processors. |
Prompt-based apps let you customize how Omi thinks and responds without writing any backend code. Simply define your prompt and Omi's behavior changes accordingly.
Everything runs on Omi's infrastructure - just write your prompt Create, test, and publish in minutesChat prompts alter Omi's conversational style and knowledge base. Create specialized versions of Omi for different contexts or expertise areas.
flowchart LR
U[User Message] --> O[Omi + Your Prompt]
O --> R[Customized Response]
style O fill:#e1f5fe
1. Your prompt is injected into the system context
2. Omi adopts the personality/expertise you defined
3. All responses reflect your customizations
The prompt persists for the entire conversation, creating a consistent experience.
Example - Fitness Coach Prompt:
You are an enthusiastic fitness coach named Alex. You:
- Provide encouraging, actionable fitness advice
- Ask about the user's goals and current fitness level
- Suggest specific exercises with proper form instructions
- Celebrate progress and motivate through setbacks
- Keep responses concise but informative
Example - Creative Writing Partner:
You are a creative writing mentor. You:
- Help brainstorm story ideas and plot structures
- Provide constructive feedback on writing samples
- Suggest techniques for improving prose
- Share examples from famous authors when relevant
- Encourage experimentation with different styles
Memory prompts analyze conversations and extract specific information based on your criteria. They run after each conversation is processed.
flowchart LR
C[Conversation Ends] --> M[Memory Processing]
M --> P[Your Prompt]
P --> E[Extracted Information]
style P fill:#e8f5e9
1. The full transcript is passed to your prompt
2. Your prompt extracts/summarizes specific information
3. The result is saved alongside the conversation
This creates structured data from unstructured conversations.
Example - Action Item Extractor:
Analyze this conversation and extract all action items.
For each action item, identify:
- The task description
- Who is responsible (if mentioned)
- Any deadline or timeframe mentioned
- Priority level (high/medium/low based on context)
Format as a numbered list. If no action items exist, respond with "No action items found."
Example - Meeting Summary:
Create a structured summary of this meeting:
1. **Attendees**: List people mentioned
2. **Main Topics**: Bullet points of key discussion areas
3. **Decisions Made**: Any conclusions or agreements
4. **Action Items**: Tasks assigned with owners
5. **Follow-ups**: Next steps or future meetings mentioned
Keep each section concise (2-3 sentences max).
You can create apps with both chat AND memory prompts for comprehensive functionality.
Example - Project Manager App:
- Chat Prompt: Acts as a project management assistant, helping plan tasks and track progress
- Memory Prompt: Extracts project updates, blockers, and timeline changes from conversations
Decide what your app will do: - **Chat only**: Personality/expertise customization - **Memory only**: Information extraction - **Both**: Complete conversational experience with data extraction Craft clear, specific prompts: - Define personality traits and expertise areas - Specify extraction criteria and output format - Include examples if helpful Before publishing, test your app extensively (see testing section below) Change visibility to public or submit for review (reviewed within 24 hours)
Open the Omi app → Settings → Enable Developer Mode Navigate to any existing memory in your app Tap the 3-dot menu (top right) → Developer Tools section Test your memory prompt against the existing conversation <iframe width="560" height="315" src="https://www.youtube.com/embed/MODjSoTMAh0" title="Testing Memory Prompts" frameborder="0" allow="accelerometer; autoplay; clipboard-write; encrypted-media; gyroscope; picture-in-picture; web-share" referrerpolicy="strict-origin-when-cross-origin" allowfullscreen ></iframe> Create your app with visibility set to private Install and enable your private app Start a conversation and verify the behavior matches expectations Refine your prompt based on the responses
Vague prompts lead to inconsistent results. Define exact behaviors, formats, and constraints. Try unusual inputs to ensure your prompt handles unexpected scenarios gracefully. Single-purpose apps perform better than apps trying to do everything. Start simple, test, and refine. Don't try to perfect the prompt on the first attempt. Avoid prompts that could generate harmful, misleading, or inappropriate content. Apps violating community guidelines will be removed.
Build webhook-based apps for external integrations Add custom tools that users can invoke in chat Publishing guidelines and review process Overview of all app types