Developer Documentation

RiskOS exposes four API layers that constrain autonomous agent decisions. Each layer operates independently but they're designed to work as a pipeline: market data flows through Axon, allocations through Tensor, verification through Myelin, and execution plans through Synapse.

Quick start

Evaluate a policy constraint in one request, or follow the complete integration loop.

"text-blue-600 dark:text-blue-400 font-semibold">curl "text-blue-600 dark:text-blue-400 font-semibold">-X "text-blue-600 dark:text-blue-400 font-semibold">POST https://api.sylaris.io/v1/policy/evaluate \
  "text-blue-600 dark:text-blue-400 font-semibold">-H "Authorization: Bearer sk_live_..." \
  "text-blue-600 dark:text-blue-400 font-semibold">-H "Content-Type: application/json" \
  "text-blue-600 dark:text-blue-400 font-semibold">-d '{
    "network": "arbitrum",
    "asset": "ETH-USD",
    "action": "buy",
    "size_usd": 50000,
    "max_slippage_bps": 15
  }'

Axon produces invariant risk artefacts — versioned model outputs that are replayable and deterministic. Five models run in parallel on GPU: fractal regime detection, GBM scenario paths, neural pressure scoring, macro regime classification, and geodesic momentum vectors.

{
  "run_id": "axn_2026_02_09_fractal_001",
  "as_of": "2026-02-09",
  "config_hash": "a1b2...c3d4",
  "model": "fractal",
  "model_version": "v0.9",
  "invariants": {
    "regime_state": "Range",
    "action": "HOLD",
    "confidence": 0.91,
    "stability_flag": true
  }
}

Tensor constrains capital allocation into opposing buy/sell baskets with hard budget limits. Submit a proposed allocation and Tensor returns whether it's within bounds — or activates refusal mode if constraints are violated.

{
  "action": "allocate",
  "portfolio_id": "pf_8a2c1e",
  "invariants": {
    "run_id": "axn_2026_02_09_fractal_001",
    "model": "fractal",
    "model_version": "v0.9",
    "as_of": "2026-02-09"
  },
  "risk_budget": 0.12,
  "universe": [
    "AAPL",
    "MSFT",
    "GOOGL",
    "AMZN",
    "NVDA",
    "JPM",
    "V",
    "UNH"
  ],
  "constraints": {
    "max_single_weight": 0.2,
    "max_sector_weight": 0.4,
    "min_positions": 5,
    "turnover_limit": 0.3
  },
  "config": {
    "model_version": "tensor.v0.4.1",
    "objective": "min_variance",
    "rebalance_frequency": "weekly"
  }
}

{
  "run_id": "tns_2026_02_09_bal_001",
  "model_version": "tensor.v0.4.1",
  "config_hash": "a7f3...c91d",
  "as_of": "2026-02-09T08:00:00Z",
  "status": "OK",
  "weights": {
    "AAPL": 0.148,
    "MSFT": 0.162,
    "GOOGL": 0.121,
    "AMZN": 0.098,
    "NVDA": 0.134,
    "JPM": 0.112,
    "V": 0.118,
    "UNH": 0.107
  },
  "constraints": [
    {
      "name": "max_single_weight",
      "status": "PASS",
      "detail": "Max 16.2% ≤ 20.0% limit"
    },
    {
      "name": "max_sector_weight",
      "status": "PASS",
      "detail": "Tech 56.3% → split across sub-sectors"
    },
    {
      "name": "min_positions",
      "status": "PASS",
      "detail": "8 positions ≥ 5 minimum"
    },
    {
      "name": "turnover_limit",
      "status": "PASS",
      "detail": "Turnover 18.4% ≤ 30.0% limit"
    }
  ]
}

Myelin scores LLM-generated trading rationale for hallucination risk. Submit the agent's reasoning alongside source signals, and Myelin returns a calibrated trust score with flagged claims.

// POST /v1/myelin/verify
// Request
{
  "text-accent-hover">"agent_reasoning": "ETH shows strong bullish momentum...",
  "text-accent-hover">"source_signals": { "text-accent-hover">"run_id": "axn_2026_03_03_001" },
  "text-accent-hover">"claim_refs": ["regime is trending", "momentum is positive"]
}

// Response
{
  "text-accent-hover">"trust_score":"text-orange-400"> 0.87,
  "text-accent-hover">"hallucination_score":"text-orange-400"> 0.02,
  "text-accent-hover">"flagged_claims": [],
  "text-accent-hover">"verified_claims":"text-orange-400"> 2,
  "text-accent-hover">"total_claims":"text-orange-400"> 2
}

Synapse produces the final trading plan by aggregating outputs from all upstream layers. The result is a bounded, auditable decision document with a cryptographic hash for institutional review.

Integration flow

Four steps from market data to policy enforcement. Every input is versioned, every output is replayable.

Monads Ledger

Every operation is tracked in the Monads ledger with audit-grade precision. Artifact generation, allocation requests, and premium features consume monads independently.