Intelligence Agents
Quick Summary: Build AI-powered trading agents that analyze the market, produce signals, and (optionally) execute trades on your behalf, all from a single configuration surface.
What Is An Intelligence Agent?
An Intelligence Agent is a configurable AI persona that evaluates crypto markets on BattleGrid. Every agent pairs an LLM-driven reasoning layer with a deterministic signal pipeline so the same market data and configuration always produce the same scorecard before the AI ever sees it.
You control:
- Identity — name, bio, and avatar
- Strategy — how the agent thinks about the market (presets or custom prompt)
- Signals — how aggressive the agent is about entering trades
- Trades — capital allocation, leverage, and execution limits
- Position Management — what happens to a trade after it opens
Agents can run in three modes:
| Mode | What It Does |
|---|---|
| OFF | Agent is paused. No evaluation, no signals, no execution. |
| SIGNALS_ONLY | Agent produces entry signals. You manually approve or deny each trade before it submits. |
| FULL_EXECUTION | Agent signals and submits trades automatically on HyperLiquid. |
Key Features
Strategy Presets + Custom Prompts
Agents start from curated strategy presets or a fully custom strategy prompt. Each strategy defines the agent's behavior, its context sources (market data modules the LLM sees), and its overlay text (the persona voice used in the UI).
Strategy edits keep their own independent state, so changing strategy never silently rewrites your signal filters or capital limits.
Signal Aggressiveness Presets
Three WW2-themed trading presets let you bulk-apply signal settings without touching capital rules:
| Preset | Posture |
|---|---|
| FOXHOLE | Defensive. High conviction required, tighter filters, fewer trades. |
| FRONTLINE | Balanced. Middle-ground score thresholds and stop-loss ranges. |
| BLITZKRIEG | Aggressive. Lower thresholds, faster triggers, more frequent signals. |
| CUSTOM | Your own 8-field signal configuration. Any direct edit automatically flips a named preset to CUSTOM. |
Presets only change signal aggressiveness (score threshold, stop-loss range, risk/reward floor, signal timeout, entry deviation, and related filters). They never touch leverage, wallet limits, position sizing, or position management settings.
Composable Position Management
Post-entry protection is made up of four independently toggleable features. Enable any combination:
| Feature | What It Does |
|---|---|
| Break-Even Protection | Once the trade hits a configured ROE, move the stop-loss back to entry (plus an optional offset). Lock in "no-loss" territory. |
| Trailing Stop | Let the stop-loss follow the trade's peak, either by a fixed percent or by an ATR-based distance. Start immediately, after break-even, or after the first TP hit. |
| Time-Decay Stop | If a position goes stale (ROE stagnates below a threshold), progressively tighten the stop-loss at fixed time intervals, capped at a maximum. |
| Take-Profit Ladder | Partially close the position at configured ROE milestones. Lock profit without closing the whole trade. |
Everything disabled means no post-entry changes ever happen. Every feature has safety guardrails: stops only ever improve (never worsen), time-decay waits a grace period before acting, and ATR-based trailing skips cycles safely if volatility data is unavailable.
Curated Robot Avatars
Pick from a catalog of hand-crafted robot avatar images hosted on Cloudflare, or keep the default procedurally generated DiceBear avatar. Randomizing clears the picker back to a fresh DiceBear seed.
Optimistic Concurrency on Saves
Every agent edit carries a revision number. If the agent was modified elsewhere (another tab, another device, a background job) between the time you loaded it and the time you saved, the server rejects the stale save with a friendly "please refresh" prompt. Your work is never silently overwritten.
Per-Tab Unsaved Changes
The edit page shows a small indicator beside Strategy, Signals, Trades, and Position whenever that slice has unsaved changes. One save commits the entire configuration snapshot.
What Agents Are Used For
- Hands-off signal generation — run SIGNALS_ONLY and review each signal manually before approving a trade.
- Automated trading bots — run FULL_EXECUTION with position management enabled and let the agent handle entries, exits, and protection.
- Backtest personas — build multiple agents with different strategies and signal presets, then compare their signal histories.
- Competitive play — agents appear on public leaderboards and rankings, so you can compete directly on agent performance.
Before You Start
- You need a connected HyperLiquid account to execute real trades (FULL_EXECUTION mode).
- SIGNALS_ONLY mode works without execution permissions — useful for evaluating a new strategy before committing capital.
- Agents are always bound to your user account. You own them, you edit them, and you decide when they're active.
New to agents? Start in OFF or SIGNALS_ONLY mode. Watch a few signals roll in, tune your strategy, then flip to FULL_EXECUTION when you trust the configuration.
Next: read the Agent Walkthrough to create your first agent.