Skip to main content

Documentation Index

Fetch the complete documentation index at: https://docs.nusomi.com/llms.txt

Use this file to discover all available pages before exploring further.

Salesforce is one of the most common Nusomi recording surfaces — it’s where the work that “should have been an API call” actually happens. This recipe covers a typical lead-routing workflow.

​
Setup

Enable browser capture on *.salesforce.com and any custom Lightning communities: 
# nusomi.workflow.yaml
workflow: salesforce_lead_route
capture:
  browser:
    domains:
      - "*.salesforce.com"
      - "*.force.com"
      - "*.lightning.force.com"
    record_screenshots: true
The DOM is the source of truth for Salesforce — Lightning’s component IDs are stable enough for exact replay, but the screenshots help training and human review.

​
Recording a workflow

The operator opens a lead, runs through scoring, assigns the territory, and submits. As they work, Nusomi captures every click, field edit, and validation error. 
import { Nusomi } from "@nusomi/sdk";

const nusomi = new Nusomi({ apiKey: process.env.NUSOMI_API_KEY });

// The desktop / browser extension picks up the workflow trigger automatically
// when the operator opens a lead URL matching the configured rule.
// Or, drive it explicitly:

const session = await nusomi.sessions.create({
  workflow: "salesforce_lead_route",
  metadata: {
    lead_id: "00Q5g000003ABCD",
    routing_strategy: "territory_v2",
  },
});

await session.start();
// ... operator works ...
await session.tag("territory_assigned", { territory: "EMEA-North" });
// ... operator submits ...
await session.tag("submitted_for_approval");
await session.stop();

​
Replaying for the next lead

When the next lead matches the same shape, Nusomi can replay the path: 
// Find a recent successful session for the same routing strategy
const [exemplar] = await nusomi.sessions.list({
  workflow: "salesforce_lead_route",
  outcome: "success",
  metadata: { routing_strategy: "territory_v2" },
  limit: 1,
});

// Guided replay against the new lead
const replay = await nusomi.replay.guided(exemplar.id, {
  sandbox: "live",
  metadata: { lead_id: "00Q5g000003WXYZ" },
});

await replay.run();
Guided replay handles the cases exact replay can’t: a new validation rule, a renamed picklist value, a Lightning component that moved.

​
Recovery after a validation error

Salesforce loves to throw inline validation errors at submit. When that happens: 
nusomi.events.on("validation_error", async (ev) => {
  if (ev.session.workflow !== "salesforce_lead_route") return;

  const recovery = await nusomi.recovery.from(ev.session_id, {
    mode: "guided",
    max_attempts: 2,
  });

  await recovery.run();
});
Recovery resumes at the frame just before the validation error, with the form’s prior state restored.

​
Training from successful runs

Once you have a few hundred recorded runs, export them as a training set: 
await nusomi.exports.create({
  workflow: "salesforce_lead_route",
  filter: { outcome: "success", since: "60d" },
  format: "webdataset",
  destination: {
    kind: "s3",
    bucket: "acme-training",
    prefix: "salesforce/lead_route/v1/",
  },
  tag: "salesforce_lead_route@v1",
});
The dataset contains a frame for every action the operator took, paired with the action itself. Hand it to your model training pipeline.

​
Tips

  • Mask PII at the field level. Salesforce contact records contain names, phone numbers, and emails. Either turn on the workspace email and phone_intl patterns or right-click-redact the sensitive fields.
  • Tag intent, not implementation. Tag territory_assigned instead of clicked-button-7. Tags are how the memory graph clusters runs.
  • Keep one workflow slug per intent. “Lead routing” is a workflow. “Lead routing for EMEA” is metadata, not a separate workflow.