How to Talk to the BlaveClaw Agent

The agent is a quantitative trading AI — not a chatbot. Give it information, not questions.

What the agent actually does

The BlaveClaw agent runs on your dedicated server and responds via Telegram. It can:

Write strategy code based on your description
Run backtests and show you the results (Sharpe, MDD, equity chart)
Scan parameter combinations and find a robust set
Fetch live market data from the Blave API (indicators, klines, alpha signals)
Deploy a strategy live with a scheduled cron job
Query your exchange account (balances, open positions)

It will not: guarantee strategy performance, give financial advice, or execute trades directly during a conversation (trades only happen through deployed scheduled strategies).

The most useful thing you can tell it

The agent can handle ambiguity, but clear input produces better output faster. The minimum useful description includes:

What	Example
Asset / symbol	BTCUSDT, ETHUSDT, 2330 (TSMC), CL (WTI crude)
Time interval	1h, 4h, 1d, 5min
Signal idea	"SMA crossover", "RSI < 30 go long", "use Taker Intensity"
Market side	Spot or futures/perpetual

Example message that works well:

幫我寫一個 BTCUSDT 1h 的策略，用 SMA 交叉訊號，
用合約，從 2022 年回測。

The agent will fill in the gaps: it will choose reasonable default parameters, suggest a warmup period, run the backtest, and show you the results. You then iterate from there.

Using Blave alpha indicators

If you want to use Blave's proprietary indicators (Taker Intensity, Holder Concentration, Whale Hunter, etc.), just name them. The agent knows how to fetch and use them. Examples:

"用多空力道（Taker Intensity）24h 做門檻策略，BTCUSDT 5min"

"用籌碼集中度過濾，只有 HC 為正時才進場"

"結合 SMA 訊號和巨鯨警報，兩個都看好才做多"

Asking for a parameter scan

After the agent writes a strategy and runs the first backtest, ask it to scan parameters. This finds a more robust combination than the default:

"幫我掃描 SMA fast 和 slow 的參數"

"scan the entry threshold and exit threshold"

The agent will run a grid search, generate a Sharpe heatmap, and recommend parameters from a plateau region (not just the single peak).

Iterating on a strategy

You don't have to get it right in one message. Common iteration patterns:

After seeing the backtest: "The MDD is too high, can we add a volatility filter?" or "Sharpe is low — try a different signal"
After the parameter scan: "Use these parameters and run the full backtest again"
After reviewing results: "OK, I want to deploy this live on Binance futures"

What the agent will always ask before deploying

The agent will never deploy a live strategy without explicit confirmation. Regardless of what you say ("try it live", "deploy it"), it will always ask:

Do you want to deploy this live? — you must reply YES
Spot or futures/perpetual?
Align current positions? (reconcile open positions on first run)
How much capital to allocate (account_value) and what's your risk tolerance (target_vol_pct)

This is by design. Deploying live has real financial consequences — the confirmation step exists to prevent mistakes.

← Back to Learn