Autonomous AI agents on ICP that coordinate to maintain fresh, up-to-date knowledge bases - powered by blockchain-native timers and direct HTTP outcalls.
Built to demonstrate how ICP's unique features enable truly autonomous AI systems that would be impossible or prohibitively expensive on other blockchains.
ResearchOS is a self-updating news intelligence system where AI agents work together autonomously:
- Ask for news on any topic → Get AI-summarized results from Groq LLM
- Enable tracking → Topic auto-updates every 20 minutes forever
- Build knowledge over time → Creates verifiable history of how news evolves
- Runs autonomously → No external triggers, servers, or maintenance needed
-
Groq AI News Agent
- Makes HTTP outcalls directly from blockchain to AI APIs
- No oracles needed (would cost $100s/day on Ethereum)
- Stores AI responses immutably on-chain
-
Timer Agent (The Game Changer)
- Runs every 20 minutes WITHOUT any external triggers
- Completely autonomous - no Chainlink Keepers needed
- Creates a truly unstoppable news aggregator
-
Storage Agent
- 48GB of persistent memory (vs KB on Ethereum)
- Maintains complete historical record
- Data survives upgrades - true permanent storage
-
Coordination Layer
- Agents share memory directly - no message passing
- Zero coordination overhead
- Seamless multi-agent collaboration
| Platform | Daily Cost | Features |
|---|---|---|
| ResearchOS on ICP | ~$0.01 | ✅ Autonomous timers ✅ Direct API calls ✅ Unlimited storage ✅ No gas per update |
| Ethereum + Chainlink | ~$500+ | ❌ Need oracles ($50/call) ❌ Need Keepers ($10/trigger) ❌ Storage extremely expensive ❌ Gas for every update |
| Traditional Server | ~$5-10 | ❌ Single point of failure ❌ Can be censored ❌ Requires maintenance ❌ Not verifiable |
1. User: "Tell me the scoop on Lagos tech scene"
↓
2. Groq AI Agent fetches & summarizes latest news
↓
3. Storage Agent caches result with timestamp
↓
4. Timer Agent adds to 20-minute update cycle
↓
5. Forever: Autonomous updates without human intervention
↓
6. User can view history: See how news evolved over time
- HTTP Outcalls: Direct blockchain→AI communication (impossible on ETH/SOL)
- Timers: Self-executing code every 20 minutes (unique to ICP)
- Stable Memory: 48GB storage at ~$0.50/GB/year (vs $1M+/GB on Ethereum)
- Low Cost: Entire system runs for pennies per day
Run the complete development and user journey:
chmod +x demo_user_journey.sh
./demo_user_journey.shWhat happens in the demo:
- Deploys autonomous news network
- User searches for "Lagos news" → Gets AI summary
- Enables tracking → Automatic updates start
- Shows timer agent updating all topics
- Displays news history over time
- Runtime: 5-7 minutes
┌─────────────────────────────────────────────────────────────┐
│ ResearchOS Canister │
├─────────────────────────────────────────────────────────────┤
│ News Agent: Queries Groq AI for summaries │
│ Timer Agent: Autonomous 20-min update cycles │
│ Storage Agent: Maintains historical knowledge base │
│ Tracking Agent: Manages monitored topics │
├─────────────────────────────────────────────────────────────┤
│ Cost: ~$0.01/day | Uptime: 100% | Censorship: Impossible │
└─────────────────────────────────────────────────────────────┘
- DFX SDK 0.15.0+
- Rust (latest stable)
- Node.js 16+ (for UI)
git clone https://github.com/yourusername/researchos
cd researchos
dfx start --clean
./demo_user_journey.sh # Compiles, deploys, and launches UI# Get news with tracking
dfx canister call research_ai_simple_backend get_news '(record {
topic = "AI breakthroughs";
enable_tracking = true
})'
# View all tracked topics
dfx canister call research_ai_simple_backend get_tracked_topics
# Check news history
dfx canister call research_ai_simple_backend get_news_history '("AI breakthroughs", opt 5)'
# Force immediate update
dfx canister call research_ai_simple_backend trigger_update_cycle- Censorship-resistant news archive
- Verifiable update history
- No single point of failure
track_topic("Bitcoin price analysis");
track_topic("S&P 500 market sentiment");
track_topic("Gold futures outlook");
// Updates every 20 minutes, forever- Track scientific breakthroughs
- Monitor paper publications
- Build citation networks over time
track_topic("OpenAI announcements");
track_topic("Google AI updates");
track_topic("Anthropic news");
// Autonomous competitor monitoringEthereum: Would need Chainlink oracles ($50/API call) + Keepers ($10/update) = $1000s/day
Solana: No HTTP outcalls, no timers, would timeout on AI responses
Traditional Backend: Single point of failure, censorable, requires maintenance
ICP: Direct API calls + autonomous timers + cheap storage = $0.01/day
- Multi-LLM Integration: Aggregate from GPT-4, Claude, Gemini
- Smart Synthesis: AI agents that cross-reference and validate
- Specialized Verticals: Financial research, academic papers, patent monitoring
- Advanced UI: Real-time dashboards and alerts
ResearchOS demonstrates a new paradigm: Autonomous AI agents that maintain living knowledge bases on the blockchain.
No servers. No maintenance. No downtime. Just intelligent agents working 24/7 to keep information fresh and accessible.
Innovation: First truly autonomous AI-powered research system using ICP's unique features
Technical Achievement: Combined HTTP outcalls + timers + stable memory for unprecedented functionality
Cost Efficiency: Runs for ~$0.01/day vs $1000s on other platforms
Canister ID: [To be deployed]
License: Apache 2.0
ResearchOS: Where AI agents live on the blockchain and work for you forever.