🎯 The problem we're solving

Not everything on our phone needs our eyes, mind, ears. But we still spend 3-5 hours daily on our phone. Most of it is repetitive, low judgement, like checking the same people, apps, updates, notification, calls.

un-app learns these patterns and acts on your behalf. No prompts. No manual work. Just done.

un-app builds YOUR AI that evolves in your phone based on your patterns, contexts, preferences in observer, helper and doer mode, to work based on your autonomy comfort.

📊 Why this matters

We surveyed 958 people and found: Most spend 4+ hours daily on repetitive phone actions - low judgement tasks, distraction calls, blinking notifications, FOMO of DND, manual focus preference setup, automating rules.

This isn't a productivity problem - it's a behavioral pattern problem.

🧠 Our approach: Behavioral memory

Instead of building another AI that answers questions, we're building an AI that learns YOUR patterns.

Check NASDAQ Monday evening? Your AI pre-fetches before you ask. Check milk availability Tuesday morning? Your AI suggests it at 8:55am.

We start with content (the wedge). We expand to cabs, food, commerce.

Observer mode → Helper mode → Doer mode. You control the autonomy.

🧠 What is Behavioral Graph

Your Behavioral Graph is the unique web of patterns that makes YOUR AI truly yours. It learns who you track when you check how often you engage and predicts what you'll need next.

• Temporal patterns - Daily 9am checks weekly Friday reviews
• Entity preferences - Tech founders vs sports players vs content creators
• Sequence learning - Checking Peter Thiel then Elon then Naval = investor mode
• Predictive accuracy - After learning your graph we anticipate needs before you ask

🤖 What agents are running

When you use un-app the following AI agents work in the background

• Pre-Fetch Agent - Runs every 60 seconds monitoring your patterns and autonomously fetching content BEFORE you ask
• Intent extraction agent - Understands your natural language queries and extracts structured intent
• Behavioral memory agent - Learns your patterns who you track when you check which sources you prefer and predicts your next query
• Content retrieval agent - Searches YouTube and the web to find real-time updates from people or channels you care about
• Pattern detection agent - Analyzes your usage over time to detect temporal rhythms preferred entities and behavioral sequences

⚙️ Technical implementation

What's automated

• Natural language processing for intent extraction
• Real-time YouTube content fetching via API
• Web search for content across platforms
• Pattern learning and behavioral prediction algorithms
• Autonomous pre-fetching based on learned patterns

Current limitations

• LinkedIn or X - No direct API access available. We use web search to find real posts
• Memory - Stored locally in your browser for this beta. Full product will have cloud sync
• Agent runs client-side - Future versions will have backend agent infrastructure

💡 How to use it

To see behavioral memory in action:

Try 2-3 related queries on the same topic (e.g., "Lex Fridman", "Naval Ravikant", "Paul Graham"). The system will detect the pattern and start learning your behavioral preferences.

Example queries:

• "Latest video from Lex Fridman"
• "New uploads from Fireship"
• "Recent content from Y Combinator"
• "Did Naval Ravikant post anything"
• "Show me Peter Thiel updates"
• "Sundar Pichai recent post"

What to expect: After 2-3 queries, scroll down to see "Behavioral Memory" section showing detected patterns. The autonomous agent runs every 60 seconds in the background, learning and pre-fetching based on your patterns.

🔒 Privacy and transparency

We see patterns not data

• All memory stored locally in YOUR browser only for this beta
• No server-side storage of personal data in beta version
• Toggle remember or don't remember anytime
• Clear all memory with one click
• Inspect exactly what the system knows about you
• Agent activity is transparent - see exactly what it does in the activity log