Kata ⚔️

A mobile-first autonomous perpetual futures dApp that makes quantitative crypto investing effortless and safe for everyone.

Kata puts an institutional-grade quantitative hedge fund in your pocket. Trading perpetuals is notoriously hostile for everyday users—over 90% lose capital to emotional trading, liquidation traps, and complex orderbooks. Kata changes this: connect your wallet, choose your risk comfort, and let an autonomous, self-improving pilot manage execution 24/7.

Most "AI trading bots" are simply language models that print reckless opinions. Kata’s real edge is everything downstream of the signal: calibrated confidence floors, walk-forward expected-value (EV) vetoes, idempotent order execution, and a 3-tier closed-loop learning engine that measurably shifts parameters as real trade outcomes land.

---

The Core Differentiator: Downstream Risk Gating

Kata's edge is capital preservation. Before any order is dispatched, it must survive rigorous statistical risk gates:

1. Calibrated Confidence Floor: Signals below a statistically backed threshold (e.g. 60%) are discarded before entry.
2. Expected Value (EV) Veto: Uses a purged walk-forward logistic policy trained on historical outcomes. If the modeled net expectancy is negative ($P(\text{win}) \cdot \text{gain} - P(\text{loss}) \cdot \text{loss} \le 0$), the signal is hard-vetoed.
3. The Downstream Veto Feed ("Why I Didn't Trade"): Rather than hiding rejections, Kata transparently logs every blocked setup with its mathematical justification—saving users from unneeded drawdowns.

---

3-Tier Closed-Loop Learning Architecture

Kata does not remain static; the loop is the product:

┌────────────────────────────────────────────────────────────────────────┐
│                        KATA LEARNING CYCLES                            │
└────────────────────────────────────────────────────────────────────────┘
  Level 1: Real-Time Feedback (Every 30s)
  └── Tracks active positions, MFE/MAE excursions, and dynamic trailing stops.

  Level 2: Pattern Matrix Clustering (Every 15m)
  └── Evaluates 24+ market archetypes (RSI states, volatility regimes, breakout vs mean rev)
      to adjust confidence multipliers and position sizes.

  Level 3: Meta-Learning Optimization (Daily)
  └── Retrains walk-forward policy models and recalibrates global confidence floors.

---

Venue-Agnostic Execution Engine

The agent depends strictly on the PerpVenue abstraction, never on a concrete exchange:

kata/venues/
├── types.py         # Venue-neutral Quote, Position, OrderRequest, OrderResult
├── base.py          # The PerpVenue abstract contract
├── hyperliquid.py   # Hyperliquid L1 adapter (live delegated execution)
├── drift.py         # Drift Protocol / Solana native adapter
└── registry.py      # Dynamic venue resolution (get_venue)

The same signal pipeline, risk engine, and learning loop operate unchanged across venues. Adding a new DEX requires an adapter, not a rewrite.

---

Web & Mobile Cockpit

Kata includes a high-performance web cockpit built with an obsidian cybernetic aesthetic:

• Live Telemetry & Pilot Status: Real-time market regime detection and active confidence floors.
• Agent Brain (ReAct Stream): Live streaming thought feed showing market scanning, thesis synthesis, and order fills.
• Downstream Veto Feed: Transparent ledger of bad setups the agent saved you from.
• Public Verifiable Track Record: Ledger of closed trades with on-chain verification links.
• Non-Custodial Allocation Console: 1-tap deposit, leverage sliders (1x–10x), and emergency pause controls.

---

Quickstart & Local Development

1. Installation

git clone https://github.com/ayomidefagboyo/kata.git
cd kata
pip install -r requirements.txt
cp .env.example .env   # Configure database, venue, and LLM credentials

2. Run the Web Cockpit & API

python start.py web

Visit http://localhost:8000/ to view the live Kata Cockpit.

3. Run the Full Stack (Web + Background Learning Loop)

python start.py all

4. Run via Docker Compose

docker compose up -d --build

---

Architecture Overview

• kata/venues/: Exchange-agnostic perp contract and adapters (Hyperliquid L1, Solana Drift).
• kata/agents/: Autonomous ReAct agent implementations (Yuki perp pilot, AgentManager).
• kata/services/: Multi-level learning engine, offline walk-forward policy, and capital allocation.
• kata/workers/: Continuous background pipelines (real-time learning, market signals scanner).
• kata/api/: FastAPI REST & WebSocket endpoints (/ws/{user_id}).
• kata/sql/: Consolidated database schema (schema.sql) for PostgreSQL / Supabase.
• client/: Lightweight, high-conviction mobile-first trading cockpit.

---

License

Proprietary. All rights reserved.