Skip to content

chat_tips.mdx

Latest commit

 

History

History
294 lines (245 loc) · 10.8 KB

chat_tips.mdx

File metadata and controls

294 lines (245 loc) · 10.8 KB
Raw
title Chat Tips & Best Practices
icon message-dots
description Learn how to get the best results from Omi chat. Discover prompting strategies, example questions, and tips based on how the system works internally.

How Your Questions Get Processed

When you ask Omi a question, it goes through an intelligent routing system that determines the best way to answer:

flowchart LR
    Q[Your Question] --> R{Router}
    R -->|Simple| NC[Fast Response]
    R -->|Context Needed| A[22+ Tools]
    R -->|Persona App| P[Character Response]
Loading
Understanding these paths helps you frame questions that get the richest, most relevant answers. **For:** Greetings, general advice, brainstorming 
Questions like "Hi!", "What's a good book to read?", or "Tips for productivity" get fast, direct responses without searching your memories.

**Examples:**
- "Hello!"
- "What's a good productivity tip?"
- "Help me brainstorm ideas for..."
**For:** Personal data, temporal queries, integration lookups 
Questions like "What did I discuss yesterday?", "Show my calendar", or "What are my goals?" trigger the full agentic system with access to 22+ tools.

**Examples:**
- "What did I discuss yesterday?"
- "Show my calendar for tomorrow"
- "What are my goals?"
**For:** Character-based apps 
If you're chatting with a persona app (like "Ask Einstein"), responses come from that character's perspective.

**Examples:**
- Asking Einstein about physics
- Getting advice from a custom coach persona

Prompting Strategies

Omi excels at temporal queries. The system converts relative time references to precise date ranges. 
<Tip>
Use natural time expressions - the system understands them perfectly.
</Tip>

**Great examples:**
- "What did I discuss yesterday?"
- "Summarize my last 3 days"
- "What happened this morning?"
- "Conversations from last week"
- "What did I talk about on Monday?"

**Less effective:**
- Vague timeframes without anchors ("a while ago")
- Very broad ranges ("everything from this year")
Omi uses **semantic search** - it finds conversations by meaning, not just keywords. Ask naturally! 
**Great examples:**
- "What have I discussed about my health?"
- "Conversations about career growth"
- "Times I talked about feeling stressed"
- "Discussions involving my manager"
- "What have I said about the project launch?"

**Why it works:** The system embeds your question and finds conversations with similar meaning, even if they use different words.
Omi builds a knowledge base of facts about you across all conversations. Tap into it! 
**Great examples:**
- "What do you know about me?"
- "What are my goals?"
- "Who do I talk to most?"
- "What are my preferences?"
- "What hobbies have I mentioned?"

**Why it works:** This triggers the memories system, which stores structured facts extracted from your conversations.
If you've connected services, ask about them directly. 
<CardGroup cols={2}>
  <Card title="Google Calendar" icon="calendar">
    - "What meetings do I have today?"
    - "When is my next meeting with Sarah?"
    - "Schedule a meeting with John at 3 PM tomorrow"
  </Card>
  <Card title="Gmail" icon="envelope">
    - "Show my recent emails about the project"
    - "Emails from John this week"
  </Card>
  <Card title="Whoop (Health)" icon="heart-pulse">
    - "How was my sleep last night?"
    - "What's my recovery score today?"
    - "Show my workouts this week"
  </Card>
  <Card title="GitHub" icon="github">
    - "My open pull requests"
    - "Issues assigned to me"
  </Card>
</CardGroup>

Example Questions by Category

Category Great Questions Why They Work
Time-Based "What did I do yesterday?", "Summarize my week" Clear time bounds trigger precise date filtering
Topic Search "What have I said about AI?", "My health discussions" Semantic search finds conceptually related conversations
People "Conversations with John", "What did Sarah mention?" People filter in search + vector similarity
Action Items "What tasks do I have?", "What do I need to do?" Direct access to action item system
Personal Facts "What's my favorite food?", "Where do I work?" Memory retrieval of stored facts
Analysis "How productive was I this month?", "Patterns in my week" Multi-tool orchestration for comprehensive answers
Calendar "My meetings tomorrow", "When am I free this week?" Calendar integration for schedule awareness

Pro Tips

These advanced strategies are based on how the system is designed internally. Narrow results by specifying both when and what: - "What did I discuss about the project **last week**?" - "Conversations about health **this month**" - "What did John say **yesterday**?" The system remembers your last 10 messages: - First: "What did I discuss yesterday?" - Follow-up: "Tell me more about the second one" - Follow-up: "What action items came from that?" Skip pleasantries when you need information: - **Better:** "My tasks for today" - **Slower:** "Hey, could you maybe show me..." Relative dates work better than absolute: - **Better:** "yesterday", "last 3 days", "this morning" - **Slower:** "January 15th, 2024" Trigger comprehensive retrieval: - "Summarize my week" - "Overview of yesterday's conversations" - "Key themes from last month" When you want thorough results: - "Show me my last 10 conversations" - "What are my top 5 priorities?"

What NOT to Ask

Some question patterns don't work well due to system design.
Avoid Why Better Alternative
"Tell me everything" Too broad, hits context limits "Summarize my last week"
"Search all my data" No specific intent to optimize for "What topics do I discuss most?"
Extremely old queries Vectors may not exist for old conversations Focus on recent conversations
Breaking persona character Confuses persona apps Stay in character with persona apps
Questions about other users Privacy boundaries Focus on your own data

Integration-Specific Tips

Best practices for calendar queries: 
| Action | Example Query |
|--------|---------------|
| Check schedule | "What meetings do I have today?" or "Am I free at 3 PM?" |
| Find meetings | "When is my next meeting with [name]?" |
| Create events | "Schedule a meeting with John tomorrow at 2 PM about project review" |
| Modify events | "Move my 3 PM meeting to 4 PM" |
Best practices for email queries: 
| Action | Example Query |
|--------|---------------|
| Search by sender | "Emails from John this week" |
| Search by topic | "Recent emails about the contract" |
| Combine filters | "Emails from Sarah about the project in the last 3 days" |
Best practices for health queries: 
| Data Type | Example Query |
|-----------|---------------|
| Sleep | "How was my sleep last night?" or "Sleep quality this week" |
| Recovery | "What's my recovery score today?" |
| Workouts | "Show my workouts this week" or "How much strain yesterday?" |
| Trends | "How has my sleep been this month?" |
Best practices for development queries: 
| Action | Example Query |
|--------|---------------|
| PRs | "My open pull requests" or "PRs I need to review" |
| Issues | "Issues assigned to me" or "Open issues in [repo]" |
| Create | "Create an issue for the login bug in [repo]" |
Best practices for Notion queries: 
| Action | Example Query |
|--------|---------------|
| Search | "Search my notes about [topic]" |
| Find pages | "Find my page about product roadmap" |

Understanding Citations

When Omi answers questions using your conversations, it includes numbered citations like [1] and [2]. These link to the source conversations.

  • Tap citations in the app to view the full conversation
  • Multiple citations mean information came from multiple conversations
  • No citations means the answer came from general knowledge or your stored memories (facts about you)

Quick Reference Card

Best Question Patterns:

Time-based:    "What did I [verb] [time period]?"
Topic-based:   "What have I discussed about [topic]?"
Person-based:  "Conversations with [name]" or "What did [name] say?"
Action items:  "What tasks do I have?" or "My to-dos"
Personal:      "What do you know about my [aspect]?"
Calendar:      "My meetings [time period]"
Summary:       "Summarize my [time period]"

Power Combinations:

Time + Topic:   "What did I discuss about [topic] [time period]?"
Time + Person:  "Conversations with [name] [time period]"
Topic + Person: "What did [name] say about [topic]?"
All three:      "What did [name] and I discuss about [topic] [time period]?"

Related Documentation

Technical deep dive into how the chat system works Create custom tools for your Omi apps How your conversations and memories are stored How live transcription works under the hood