Deploy Terminal
Autonomous Trading Terminal - Natural Language → DeFi Execution
🟢 Design Phase
Phase 1 - MVP Planning
🎯 Project Overview
Deploy Terminal is an autonomous trading terminal that collapses the institutional trading stack into a natural language AI interface. Users can execute complex DeFi strategies across multiple chains using simple commands.
Key Features
- Natural Language Interface: Execute trades with simple commands like "Lend my 1000 USDC on Aptos"
- Multi-Chain Support: Aptos, Solana, EVM chains with unified abstraction layer
- Intelligent Strategy Engine: Automated DCA, grid trading, yield farming, and more
- Production-Grade Safety: Multi-phase validation, simulation before execution
- Horizontal Scalability: Designed to handle 10,000+ concurrent users
📊 Current Status
Last Updated: 2026-01-29
Completed Milestones
- ✅ Architecture v1 design (3 competing architectures evaluated)
- ✅ Benchmarking simulation (20 test scenarios)
- ✅ Architecture C recommended (Workflow-First Hybrid)
In Progress
- 🔨 Architecture design review
- 🔨 GitHub Pages dashboard setup
Next Steps
- 📋 Build Haiku classifier + routing layer
- 📋 Implement 5 core workflows (lend, borrow, swap, stake, DCA)
- 📋 Aptos chain adapter (CAL)
🏗️ Architecture Approach
After evaluating three agent architecture patterns, we selected Architecture C: Workflow-First Hybrid
Why Hybrid?
- Fastest: 4.1s avg latency (workflows skip LLM entirely)
- Cheapest: $0.14 avg cost (5x cheaper than pure agent)
- Flexible: Falls back to Opus agent for complex strategies
- Optimal for production: 80% of requests are common patterns
Design Principles
- Correctness Over Speed: Multi-phase validation pipeline
- Horizontal Scalability: Stateless microservices, 10K+ users
- Chain-Agnostic Core: Unified adapter pattern for all chains
- Observable by Default: Distributed tracing, structured logging
- Graceful Degradation: Circuit breakers, fallback strategies
🚀 Roadmap
Phase 1: MVP (Month 1-2)
- Implement 5 core workflows (lending, DCA, grid, staking, swap)
- Build Haiku classifier
- Build Opus fallback agent
- Focus: Aptos chain only
Phase 2: Production (Month 3-4)
- Add 15 more workflows
- Optimize classifier accuracy
- Add chains: Solana, EVM
- Implement observability stack
Phase 3: Scale (Month 5-6)
- Orchestrator-worker hybrid for medium strategies
- Auto-generate workflows from agent behavior
- User-defined custom strategies
📈 Success Metrics
| Metric |
Target |
Current |
| Strategy Correctness |
≥95% |
Design phase |
| Avg Response Latency |
<5s |
4.1s (simulated) |
| Concurrent Users |
10,000+ |
Architecture ready |
| Supported Chains |
3+ (Aptos, Solana, EVM) |
0 (in design) |
| Predefined Workflows |
20 |
0 (planned) |