For AI Agents

Structured Context

OpenBio doesn't give your agent a wall of text. It gives typed assertions with trust scores, provenance chains, and temporal context — exactly what LLMs need to reason about biological evidence without hallucinating.

What the payload looks like

The agents.evidenceContext endpoint returns a structured JSON object — not prose. Every assertion is individually typed with a trust layer and confidence score.

json

{
  "subject": {
    "id": "james-58",
    "name": "James",
    "age": 58,
    "subjectType": "human"
  },
  "assertions": [
    {
      "id": "assert-001",
      "type": "lab_result",
      "value": 9.2,
      "unit": "g/dL",
      "code": "LOINC:718-7",
      "display": "Hemoglobin",
      "trustLayer": "source_fact",
      "confidence": 0.99,
      "temporalContext": {
        "observedAt": "2025-03-15T08:42:00Z",
        "certainty": "definite"
      },
      "provenance": {
        "source": "Quest Diagnostics",
        "protocol": "HL7v2 ORU",
        "collectedAt": "2025-03-15T08:42:00Z",
        "receivedAt": "2025-03-15T09:11:00Z"
      }
    },
    {
      "id": "assert-002",
      "type": "genomic_variant",
      "code": "EGFR_L858R",
      "display": "EGFR L858R mutation",
      "trustLayer": "normalized_fact",
      "confidence": 0.97,
      "temporalContext": {
        "observedAt": "2025-01-20T14:30:00Z",
        "certainty": "definite"
      },
      "provenance": {
        "source": "Foundation Medicine",
        "protocol": "NGS panel",
        "collectedAt": "2025-01-20T14:30:00Z",
        "receivedAt": "2025-01-21T09:00:00Z"
      }
    }
  ],
  "summary": {
    "totalCount": 847,
    "byType": {
      "lab_result": 238,
      "genomic_variant": 187,
      "imaging_finding": 136,
      "medication": 152,
      "diagnosis": 134
    },
    "byTrustLayer": {
      "source_fact": 412,
      "normalized_fact": 289,
      "derived_feature": 112,
      "model_output": 34
    }
  }
}

Retrieving context

Use the SDK to retrieve structured context. Filter by trust layer to control the epistemic quality of what your agent sees.

typescript

const context = await client.agents.evidenceContext({
  subjectId: 'james-58',
  format: 'structured',
  maxAssertions: 50,
  trustLayers: ['source_fact', 'normalized_fact'],
  // Optionally filter to specific evidence types
  types: ['lab_result', 'genomic_variant', 'diagnosis'],
});

// context.assertions — typed array, ready for LLM input
// context.summary   — aggregate stats for grounding
// context.subject   — subject metadata

Why structured matters

LLMs hallucinate when they can't distinguish between a measured fact and a model prediction. OpenBio's typed payload makes that distinction explicit and machine-readable.

Without OpenBio

"Patient has hemoglobin of 9.2 g/dL, EGFR L858R mutation, and is likely responding to erlotinib treatment."

Mixed trust levels. Agent cannot tell which facts are raw measurements vs. AI predictions.

With OpenBio

{ "value": 9.2, "trustLayer": "source_fact", "confidence": 0.99 }
{ "code": "EGFR_L858R", "trustLayer": "normalized_fact", "confidence": 0.97 }
{ "type": "model_output", "confidence": 0.73 }

Each assertion carries its trust level. The agent knows exactly what to trust.