Risk limits
for autonomous crypto agents.
A deterministic policy engine that tells AI agents how much risk is allowed. 30+ real-time signals. 5-level policy. Verifiable audit trail.
v1.1 live · BTC & ETH
layer
layer
leverage: 1.5x
blocked: [ALL_IN]
layer
The problem
AI agents lack risk enforcement before execution.
No policy gate
Execution stacks handle signals, routing, custody and security. Market-aware risk permissioning is still missing.
Static limits don't adapt
Fixed caps and static guardrails behave the same in a crash and a breakout. Markets change. Your limits should too.
Autonomous ≠ uncontrolled
Agents can decide what to do. A separate policy layer should decide how much risk is allowed.
Raw models find reasons to break discipline.
RiskState prevents:
- Blowups and ruin risk
- Oversized positions in weak conditions
- Leverage when the market doesn't support it
- Overtrading in low-quality setups
- Blind risk-taking on top of unhealthy DeFi loans
How it works
From signals to permissions in seconds.
RiskState policy engine converts market state into dynamic risk permissions.
Market, on-chain, macro, derivatives, DeFi health
Normalize, composite score, regime detection
4 independent caps × quality × volatility
Max size, leverage, allowed/blocked actions
5-level policy output
API Preview
One endpoint. Full risk enforcement.
Designed as a binding risk layer for agents and execution systems.
curl -X POST https://riskstate.ai/v1/risk-state \
-H "Authorization: Bearer $RISKSTATE_API_KEY" \
-H "Content-Type: application/json" \
-d '{"asset": "BTC"}'{
"exposure_policy": {
"max_size_fraction": 0.42, // ◀
"leverage_allowed": true, // ◀
"allowed_actions": ["DCA", "LONG_SHORT_CONFIRMED"],
"blocked_actions": ["ALL_IN", "LEVERAGE_GT_2X"]
},
"tactical_state": "LEAN BULL", // ◀
"market_regime": "TREND",
"policy_level": 4,
"binding_constraint": {
"source": "MACRO", // ◀
"reason_codes": ["MACRO_NEUTRAL", "COUPLING_NORMAL"]
},
"risk_flags": {
"structural_blockers": [],
"context_risks": ["HIGH_COUPLING", "TREND_NOT_CONFIRMED"]
}
}Exposure Policy
BindingHow much can I risk?
- One number:
max_size_fraction. Your agent’s hard ceiling on position size. - Enum actions (
DCA,WAIT,LEVERAGE_2X) your agent reads directly. No parsing. structural_blockersempty = go. Not empty = halt. Zero ambiguity.
Market Intelligence
ContextualBut why? Every number is auditable.
- Tactical + structural + regime in 3 fields. Context, not a wall of text.
binding_constrainttells your agent exactly why it’s limited — debuggable.include_details=trueunlocks: squeeze detector, whale pressure, trend decomposition.
DeFi-Aware
Unified RiskYour health factor changes your position limit in real-time.
- Pass a wallet address. Health factor feeds directly into
max_size_fraction. - DeFi lending risk × market risk = one position limit.
- Protocol-aware: Spark and Aave V3 with per-collateral thresholds.
Integration
Pre-trade risk gate for any system
Three ways to connect RiskState to agents, bots, and execution stacks.
REST API
LiveStandard HTTP endpoint. Any language, any framework.
import requests
response = requests.post(
"https://riskstate.ai/v1/risk-state",
headers={"Authorization": "Bearer $TOKEN"},
json={"asset": "BTC"}
)
policy = response.json()
max_size = policy["exposure_policy"]["max_size_fraction"]SKILL.md
LiveDiscovery standard for autonomous agent ecosystems.
# SKILL.md
name: riskstate
version: 1.1.1
category: risk-management
auth: bearer-token
endpoint: POST /v1/risk-state
assets: [BTC, ETH]
refresh: 60s cache, recommend 5min pollingMCP Server
LiveNative integration for Claude Desktop, Claude Code, and MCP-compatible clients.
// claude_desktop_config.json
{
"mcpServers": {
"riskstate": {
"command": "npx",
"args": ["-p", "@riskstate/mcp-server", "riskstate-mcp"],
"env": {
"RISKSTATE_API_KEY": "your-api-key"
}
}
}
}Trust
Why RiskState is trustworthy.
Built for systems that require consistent, auditable risk decisions.
Same inputs, same outputs, always. SHA-256 policy hashes on every response. No stochastic decisions, no hidden state. The policy function is pure.
External mathematical review of all scoring functions and normalization curves. Smooth sigmoids, logistic policy caps, Fisher z-transform confidence intervals. Every formula verified for continuity and boundary behavior.
Real market snapshots with forward returns across bull, bear, and sideways regimes. Quintile monotonicity testing, autocorrelation-adjusted sample sizes. Not backtested — continuously validated.
A human-readable view of how RiskState reaches risk decisions.
RiskState VisualizerRiskState doesn’t execute trades, doesn’t generate signals, doesn’t hold positions. A pure risk engine with no skin in the game beyond your outcomes.
Request early access.
Free during beta. API key delivered instantly.
FAQ