Oneechan watches live on-chain market data, asks a DeepSeek model to reason over it, sizes the trade with a risk engine, and executes it through the Hyperliquid SDK — paper-first, live only when you're ready.
Runs fully in paper mode out of the box — no keys required to try it.
Every decision cycle runs the same five-stage pipeline — read, reason, size, execute, record.
Pulls candles, funding and the order book from the Hyperliquid Info API, then computes SMA, EMA, RSI and ATR indicators.
Sends a compact context prompt to DeepSeek and parses a structured LONG / SHORT / CLOSE / HOLD decision.
Sizes the position from equity-at-risk to the stop, capped by max notional and max leverage.
Fills orders in simulated paper mode, or signs real market orders via the SDK in live mode.
Persists equity, positions and trade history to disk every cycle so you can track performance.
DeepSeek, Claude, GPT-4o, Gemini and more — OpenRouter, Groq, xAI, Mistral. One-line swap.
Simulated fills at mid price with a taker-fee assumption. Zero keys required to start.
Real orders on Hyperliquid testnet or mainnet behind an explicit --mode live flag.
Confidence-scaled sizing, stop & take-profit, max notional, max leverage, per-coin cooldown.
Watch any Hyperliquid perps — BTC, ETH, SOL and hundreds more.
Ships with a backtesting harness that replays real Hyperliquid candles and reports honest metrics — return, win rate, drawdown.
Everything below runs in paper mode — no API keys or funds required. Only switch to live when you're ready.
Python 3.10+ (tested on 3.14). A DeepSeek API key for the real brain. For live mode, a Hyperliquid wallet or API-wallet private key.
Create a virtual environment and install the dependencies.
# clone / enter the project cd oneechan # create a virtual environment and install python -m venv .venv source .venv/bin/activate # Windows: .venv\Scripts\activate pip install -r requirements.txt
Copy the environment template and fill in your keys. Paper mode needs none of this.
cp .env.example .env
# .env DEEPSEEK_API_KEY=sk-xxxxxxxx # DeepSeek key HYPERLIQUID_PRIVATE_KEY=0x... # only for live mode (API wallet)
Try a single decision cycle first, then let it run on a loop.
python main.py --once # one decision cycle, then exit python main.py # loop every 5 minutes python main.py --brain mock # run the pipeline with no LLM at all
Real orders. Start on testnet, and use an API wallet that can trade but never withdraw.
python main.py --mode live --network testnet python main.py --mode live --network mainnet
All settings live in config.yaml. Defaults are safe and paper-only.
| Key | Default | Meaning |
|---|---|---|
| agent.mode | paper | paper (simulated) or live (real orders) |
| agent.coins | [BTC, ETH, SOL] | Perps the agent watches |
| agent.interval | 1h | Candle interval fed to the brain |
| agent.lookback | 96 | Candles in the context window |
| agent.cadence_seconds | 300 | Seconds between decision cycles |
| agent.cooldown_seconds | 3600 | Min seconds between trades per coin |
| agent.max_position_pct | 25.0 | Max notional per position, % of equity |
| agent.max_leverage | 3 | Max effective leverage |
| agent.risk_per_trade_pct | 1.0 | Equity at risk per trade, % |
| llm.provider | deepseek | deepseek | openai | anthropic | gemini | openrouter | groq | xai | … |
| llm.model | deepseek-v4-pro | Provider model ID (claude-sonnet-5-5, gpt-4o, …) |
| hyperliquid.network | testnet | testnet or mainnet |
| hyperliquid.address | "" | Main wallet address (public) |
Oneechan speaks to every major model through one interface. Pick any provider below, set llm.provider and the matching key in .env.
The private key is only ever read from the environment. .env is gitignored and never committed.
Trade with a dedicated API wallet that can place and cancel orders but can never withdraw funds.
Nothing is signed unless you explicitly pass --mode live. Paper mode needs no key at all.
No. Paper mode simulates everything with a $10,000 virtual balance and needs zero keys or funds.
It returns a single structured signal — LONG, SHORT, CLOSE or HOLD — plus a confidence score, stop and take-profit levels, and a suggested leverage. The risk engine then turns that into a concrete position size.
No. Oneechan is educational software. Perpetual futures carry substantial risk of loss. Start in paper mode and never trade money you can't afford to lose.
Add HYPERLIQUID_PRIVATE_KEY to .env and run with --mode live. Start on --network testnet before touching mainnet.
Yes. Change agent.coins in config.yaml, or pass --coins BTC SOL ETH on the command line.
It lets you test the full pipeline — data, sizing, execution, ledger — without a DeepSeek key. It uses a simple momentum rule and is not meant for real trading.
Clone the repo, run it in paper mode, and watch DeepSeek read the Hyperliquid market — before you ever risk a cent.
Get started →