Configuration

All trading parameters in one place.

Trading Parameters

Configure via the dashboard, POST /agent/config, or edit defaults in src/strategy/default/config.ts:

Setting	Default	Description
`max_positions`	5	Maximum concurrent positions
`max_position_value`	5000	Maximum $ per position
`take_profit_pct`	10	Auto-sell at this % profit
`stop_loss_pct`	5	Auto-sell at this % loss
`risk_per_trade_pct`	0.75	Maximum account risk per trade, applied with stop distance and ATR-aware sizing
`min_sentiment_score`	0.3	Minimum sentiment to consider (0-1)
`min_signal_quality_score`	0.35	Minimum source quality, freshness, volume, and direction-agreement score before research
`min_analyst_confidence`	0.6	Minimum LLM confidence to trade (0-1)
`position_size_pct_of_cash`	15	Max % of available cash per trade before candidate score and risk-budget sizing
`entry_require_technical_data`	false	Require RSI/SMA/Bollinger data before allowing entry-timing-filtered buys
`dynamic_tp_fallback_pct`	12	Dynamic take-profit target used when ATR is unavailable
`unknown_sector_max_positions`	2	Separate concentration cap for symbols whose sector cannot be inferred

LLM Configuration

Setting	Default	Description
`llm_provider`	openai-raw	`openai-raw`, `ai-sdk`, or `cloudflare-gateway`
`llm_model`	gpt-4o-mini	Model for bulk research (cheap, fast)
`llm_analyst_model`	gpt-4o	Model for trading decisions (smarter)

Timing Parameters

Market timing uses the Alpaca clock (next_open, is_open) so it stays correct across timezones and holidays.

Setting	Default	Description
`data_poll_interval_ms`	30000	How often to fetch signals (30s)
`analyst_interval_ms`	120000	How often to run trading logic (2min)
`premarket_plan_window_minutes`	5	Generate a pre-market plan when within N minutes of `next_open`
`market_open_execute_window_minutes`	2	Execute the pre-market plan if the market is open and within this window after open

Optional Features

Setting	Default	Description
`options_enabled`	false	Enable options trading
`crypto_enabled`	false	Enable 24/7 crypto trading
`crypto_symbols`	["BTC/USD", "ETH/USD", "SOL/USD"]	Which cryptos to trade
`stale_position_enabled`	true	Auto-exit positions that lose momentum
`ticker_blacklist`	[]	Symbols to never trade (e.g., insider restrictions)

Data Source Credentials

Sentinel Settings is the preferred place to manage source credentials because it supports account rotation and connection tests. Worker secrets are useful as fallback defaults.

Setting	Secret fallback	Description
`twitter_cookie_accounts`	`TWITTER_COOKIES`	Multiple Twitter/X cookie accounts rotated for confirmation searches
`reddit_cookie_accounts`	`REDDIT_COOKIES`	Multiple Reddit cookie accounts rotated by the Reddit gatherer
`reddit_user_agent`	`REDDIT_USER_AGENT`	User-agent used for Reddit requests
`alpha_vantage_api_key`	`ALPHA_VANTAGE_API_KEY`	Enables Alpha Vantage news sentiment catalyst signals

Environment Variables

Set in .dev.vars (local) or via wrangler secret put (production):

Variable	Required	Description
`MAHORAGA_API_TOKEN`	Yes	API auth token for all endpoints. Generate with: `openssl rand -base64 48`
`ALPACA_API_KEY`	Yes	Your Alpaca API key
`ALPACA_API_SECRET`	Yes	Your Alpaca API secret
`ALPACA_PAPER`	Yes	`true` for paper trading (recommended)
`OPENAI_API_KEY`	Depends	Required if using `openai-raw` or `ai-sdk` with OpenAI
`KILL_SWITCH_SECRET`	Yes	Separate secret for emergency kill switch
`DISCORD_WEBHOOK_URL`	No	Get Discord alerts on BUY signals
`TWITTER_BEARER_TOKEN`	No	Optional fallback for Twitter confirmation if cookies are not configured
`TWITTER_COOKIES`	No	Optional single-account fallback; Sentinel config supports multiple accounts
`REDDIT_COOKIES`	No	Optional single-account fallback; Sentinel config supports multiple accounts
`ALPHA_VANTAGE_API_KEY`	No	Optional news sentiment source; can also be stored in runtime config

Safety Features

These are enforced by the PolicyEngine (via PolicyBroker for autonomous trades, or via MCP for external tools):

Feature	Description
Kill Switch	Emergency halt for all trading
Position Limits	Max positions and $ per position
Daily Loss Limit	Stops trading after 2% daily loss
Cooldown Period	30-minute pause after losses
Approval Tokens	Orders expire after 5 minutes
LLM Confidence Gate	Trades require minimum confidence
No Margin	Cash-only trading
No Shorting	Long positions only

API Endpoints

All endpoints require Authorization: Bearer $MAHORAGA_API_TOKEN

Endpoint	Description
`GET /agent/status`	Full status: account, positions, signals, config
`GET /agent/enable`	Start the trading agent
`GET /agent/disable`	Stop the trading agent
`GET /agent/config`	Get current configuration
`POST /agent/config`	Update configuration (JSON body)
`GET /agent/logs`	Recent activity logs
`GET /agent/trade-review`	Indexed trade decisions with optional R2 snapshots for analysis
`POST /agent/twitter/test`	Test Twitter/X cookie accounts from Sentinel Settings
`POST /agent/reddit/test`	Test Reddit cookie accounts from Sentinel Settings
`GET /agent/costs`	LLM usage and costs
`GET /agent/signals`	Current signal cache
`GET /agent/trigger`	Manually trigger the agent loop (testing)
`GET /agent/kill`	Emergency stop (requires `KILL_SWITCH_SECRET`)

Tuning Tips

Agent not making trades?

Lower min_analyst_confidence to 0.5
Lower min_sentiment_score to 0.2
Check max_positions — you might be at the limit
Check /agent/status for signal research results

Too many trades?

Raise min_analyst_confidence to 0.7+
Raise min_sentiment_score to 0.4+
Lower max_positions

High LLM costs?

Use gpt-4o-mini for both models (cheapest)
Increase data_poll_interval_ms to 60000+
Check /agent/costs for usage breakdown

Using a different LLM provider?

Set llm_provider to ai-sdk
Set llm_model to anthropic/claude-sonnet-4 or google/gemini-2.5-flash
Add the corresponding API key (e.g., ANTHROPIC_API_KEY)