From an agent's decision to evidence, in four stages.

import stratl

with stratl.decision(
    agent="support-bot",
    subject=("order", "48213"),
    agent_version="2026.09.1",
) as d:
    d.input(ticket)
    d.model("openai", "gpt-5", prompt, response)
    d.tool("orders.read", {"order_id": "48213"}, order)
    d.policy("refunds_above_100_require_approval", version=7, result="human_required")
    d.approval(required=True, approver="priya@example.com", authority="support_lead",
               decision="approved", identity_provider="okta")
    d.action("refunds.issue", target="order/48213", result="executed")
    d.outcome(adverse=False, category="refund")

Whatever you pass in is fingerprinted on your side. The record carries the fingerprints; the originals are uploaded to the custody bucket so they can be re-hashed later. If Stratl cannot be reached, the record is spooled to disk and sent on the next run.

import { Client, decision } from "@stratl/sdk";

const client = new Client(); // reads STRATL_API_URL and STRATL_API_KEY

await decision("support-bot", ["order", "48213"], async (d) => {
  await d.input(ticket);
  await d.model("openai", "gpt-5", prompt, response);
  await d.tool("orders.read", { order_id: "48213" }, order);
  await d.policy("refunds_above_100_require_approval", { version: 7, result: "human_required" });
  await d.approval({ required: true, approver: "priya@example.com", authority: "support_lead",
                     decision: "approved", identityProvider: "okta" });
  await d.action("refunds.issue", { target: "order/48213", result: "executed" });
  await d.outcome({ adverse: false, category: "refund" });
}, client, { agentVersion: "2026.09.1" });

Same rules as Python: content is fingerprinted on your side, the record carries the fingerprints, and the originals are uploaded to the custody bucket. The two SDKs produce the same fingerprint for the same content. Under Node, an unreachable API means the record is spooled to disk and sent later.

exporters:
  otlphttp/stratl:
    traces_endpoint: https://api.stratl.ai/v1/ingest/otlp/v1/traces
    encoding: json
    headers:
      Authorization: Bearer ${env:STRATL_API_KEY}

service:
  pipelines:
    traces:
      exporters: [otlphttp/stratl]

Point a collector at Stratl and one trace becomes one decision record. Set stratl.subject.type and stratl.subject.id on any span so the record is filed under the business subject rather than the trace id.

curl -X POST https://api.stratl.ai/v1/ingest/srf \
  -H "Authorization: Bearer $STRATL_API_KEY" \
  -H "Content-Type: application/json" \
  -d @decision.json

Any language can post a Stratl Record Format document. Stratl computes the hash, signs it and links it into the workspace chain. Posting the same decision_id twice is safe.