Agent-ready assurance runtime

One runtime. Every interface.

Humans, pipelines, and AI agents are all clients of the same public API. The same analysis submitted through web, API, SDK, CLI, or MCP resolves to the same canonical AnalysisSpec hash — and produces the same assured evidence object.

Python SDK

result = genomarker.run(
    analysis="differential_expression",
    spec=spec,                       # canonical, deterministically hashed
    assurance_profile="GM-BIOMARKER-DE-1",
    idempotency_key="trial-042-de-v3")

result.evidence_id        # "GM-2026-08-14-8F29E1"
result.findings           # structured, machine-readable
result.replay.outcome     # REPRODUCED_EQUIVALENT
result.preservation.grade # "A"

MCP surface — assurance only

MCP tools

list_analysis_types
get_analysis_schema
list_assurance_profiles
validate_analysis_spec
submit_analysis · get_result
get_assurance_findings
get_evidence · verify_evidence
replay_evidence
get_sap · validate_against_sap

Deliberately narrow. Genomarker’s job is to assure computation — agents can gather scientific context elsewhere.

Actor provenance

Actor provenance JSON

{
  "actor_type": "agent",
  "interface": "mcp",
  "client": "helix-sci-agent",
  "model": "optional-identifier"
}