LinkedIn Jobs Scraper | Remove Duplicates | AI Enrichment

Turn public LinkedIn jobs into clean, source-linked records for recurring alerts, job boards, databases, and automations. Search by role, location, freshness, and workplace arrangement; keep the posting details the source provides; and use cross-run deduplication so scheduled searches can return only jobs not already delivered for the same alert.

Optional features can fill missing description-backed facts, translate selected fields to English, or add facts from the public company page already linked by a job. The base source data always wins over optional enrichment.

This is an independent Actor. It is not affiliated with or endorsed by LinkedIn.

What you can build

• Daily or scheduled job alerts: rerun the same search and suppress jobs already delivered within that alert scope.
• Job-board and database feeds: store stable, source-linked records with complete descriptions when available, structured requirements, and application details.
• Automation and AI workflows: send clean records to n8n, Make, Airtable, MCP clients, webhooks, or your own API consumer.

Ready-made integrations: MCP · n8n · Make · Airtable · Claude / Codex skill

Why use this Actor

• One stable schema: every job uses the same nomad-agent-job-v1 shape, which also works with other normalized Nomad Agent job sources.
• Rich job details: titles, companies, locations, workplace arrangements, dates, application links, requirements, skills, benefits, compensation, constraints, and complete descriptions when available.
• Source-faithful data: missing facts stay unknown, source-established empty lists stay empty, and optional enrichment never overwrites source facts.
• Precise selection: use LinkedIn search controls, strict normalized filters, physical-geography checks, and optional public company filters.
• Alert-friendly delivery: built-in cross-run deduplication can return only jobs not already delivered for the same search or alert scope.
• Integration-ready: export through Apify or use the API, MCP, n8n, Make, Airtable, webhooks, and the provided parsers.

What each result contains

Every item has exactly six top-level fields:

Field	Meaning
schemaVersion	Always nomad-agent-job-v1
identity	LinkedIn source, posting ID, and canonical job URL
data	Normalized job, company, location, employment, application, requirement, benefit, compensation, and constraint fields
custom	null, or a versioned public company-profile extension when requested and available
llm	Status and provenance for optional description-backed enrichment
raw	Complete description text and source HTML, or null when includeRaw is false

All declared fields are present. null means unknown or unavailable. [] means the source explicitly established an empty collection. Consumers should not treat those values as interchangeable.

Important accuracy rules:

• a location does not by itself prove that a job is on-site;
• only named people are returned as hiring contacts;
• the posting URL and an external application URL remain separate;
• optional AI fills only allowlisted fields that remain null after source parsing;
• raw HTML is untrusted source content and must be sanitized before rendering.

Work-arrangement evidence

workArrangements is both a LinkedIn search constraint and a normalized-field filter. The Actor keeps the two meanings separate so the search does not invent a job fact:

Evidence for a returned job	Filter decision	data.employment.workArrangements
Explicit arrangement overlaps the request	Keep	Exact source or AI-filled value
Explicit arrangement contradicts the request	Exclude	Not returned
Arrangement missing, but the job came from the matching LinkedIn f_WT request	Keep	null
Arrangement missing without matching f_WT provenance	Exclude	null

LinkedIn can apply a remote/hybrid union without identifying which member applies to an individual public job. In that case the job is eligible for the requested search, but its normalized arrangement remains unknown. Search eligibility does not turn an unknown per-job arrangement into a known fact. Description prose is not treated as source-confirmed evidence.

Input

Field	Default	Purpose
firstRunMode	false	Small evaluation mode: at most five jobs, Silver AI enrichment and English translation on, dedupe/raw/analytics off
schemaVersion	required	Must be nomad-agent-job-search-input-v1
keyword	""	Job title, skill, or role
location	""	City, region, country, or supported macro-region
linkedinSearch	omitted	Up to eight keyword/location searches in one run
postedWithin	"24h"	1h, 24h, 7d, 30d, or any
orderBy	"newest"	Final result order: newest or oldest
workArrangements	omitted	Any combination of remote, hybrid, and onsite
strictGeography	omitted	Require source-backed country, region, or city matches
filters	omitted	Versioned filters over normalized job fields
maxItems	100	Maximum returned jobs; 0 requests the bounded 1,000-item window
dedupe	enabled	Suppress jobs already delivered in the same scope
companyProfileEnrichment	false	Add source-backed facts from linked public company pages
companyFilters	omitted	Filter on available public company-profile facts
aiEnrichment	disabled	Fill selected missing facts from the complete description
translateToEnglish	false	Translate selected normalized display fields
includeRaw	true	Include complete description text and HTML
analyticsEnabled	false	Share aggregate operational analytics with the Actor owner

Unknown input fields are rejected. This prevents misspelled or retired options from silently changing the meaning of a run.

orderBy is a top-level normalized-Actor control. The separately versioned linkedinSearch extension is now nomad-agent-linkedin-search-v2 and contains only schemaVersion plus optional searches; legacy nested linkedinSearch.orderBy is rejected.

First-run mode

The Apify console prefills firstRunMode: true for a small enriched first result. The mode preserves your search terms, freshness, work-arrangement, geography, and normalized/company filters, while applying these effective settings before validation and billing:

• maxItems is capped at 5; a smaller explicit limit stays smaller;
• aiEnrichment is forced to Silver and translateToEnglish is forced on;
• cross-run deduplication is disabled so the evaluation can be repeated;
• includeRaw and anonymous analytics are disabled. Raw source content can still be used internally for requested enrichment and is then returned as top-level raw: null.

AI enrichment and translation are separate paid events in addition to the base result event. Start with a $0.10 Maximum cost per run; the cap can reduce or stop output, and the Actor's Pricing tab is authoritative for current event prices. Disable first-run mode after evaluation to use the explicit feature, output, result-limit, and delivery settings below. API calls that omit the field, or send false, retain the existing behavior.

{
  "schemaVersion": "nomad-agent-job-search-input-v1",
  "firstRunMode": true,
  "keyword": "software engineer",
  "location": "Spain",
  "postedWithin": "24h"
}

Example search

{
  "schemaVersion": "nomad-agent-job-search-input-v1",
  "keyword": "software engineer",
  "location": "Spain",
  "postedWithin": "7d",
  "workArrangements": ["remote", "hybrid"],
  "maxItems": 25,
  "dedupe": {"enabled": false, "key": ""},
  "aiEnrichment": {"enabled": false, "accuracy": "silver"},
  "translateToEnglish": false,
  "includeRaw": false,
  "analyticsEnabled": false
}

Disable cross-run deduplication for repeatable one-off searches. Leave it enabled for scheduled alerts, using a stable public key only when multiple searches should intentionally share delivery history.

Optional features

Description-backed AI enrichment

Enable aiEnrichment with a silver or gold accuracy profile. The Actor reads the complete plain-text job description and fills only supported fields that static source parsing left null. Source facts and source-established empty arrays always win. No customer model key is required, and an enrichment failure leaves the base job unchanged with a failed status in llm.

When enrichment is enabled, normalized filtering is completed against the source record plus supported fields filled by AI. This includes data.employment.workArrangements: an AI-filled match can keep a job and an AI-filled contradiction excludes it. AI provenance is recorded in llm; the same words in a description never become static source evidence when AI is disabled. Translation still happens after filtering and does not affect a match.

{"aiEnrichment": {"enabled": true, "accuracy": "silver"}}

English translation

translateToEnglish: true translates selected normalized display fields that are not already confidently English: title, classifications, domains, applicant-requirement prose, benefits, eligibility and selection text, work authorization, security clearance, and location preference. Descriptions, company names, locations, identifiers, URLs, source-raw labels, skills, qualifications, certifications, programme names, and provenance remain unchanged. No customer translation key is required.

Public company profiles

companyProfileEnrichment: true can add exact public facts from the canonical LinkedIn company page already linked by a job. Unavailable or unverifiable profiles are fail-open when no company filter is requested: the complete base job remains and custom is null. companyFilters can then select jobs by supported company fields before the final result limit is applied.

Company filtering deliberately remains fail-closed. An unavailable or unverified profile is unknown, not a verified profile whose individual fields are null. With unknownPolicy: "exclude" (the default), its job is excluded. With unknownPolicy: "abort", the run fails instead of returning a row whose company predicate could not be evaluated. A verified profile with a missing field is evaluated normally, including is_null and is_not_null.

Normalized filters

Use filters for versioned AND/OR expressions over supported normalized job fields. Filters run on source-language values before optional output translation. With AI enrichment disabled, unknown facts do not become guessed matches. With enrichment enabled, predicates over supported AI-fillable fields are finally evaluated using those enriched normalized fields; source facts still take precedence.

{
  "filters": {
    "schemaVersion": "nomad-agent-job-filter-v1",
    "expression": {
      "all": [
        {"field": "data.locations[].countryCode", "operator": "eq", "value": "ES"},
        {"field": "data.title", "operator": "not_contains", "value": "internship"}
      ]
    }
  }
}

Run with Python

from decimal import Decimal

from apify_client import ApifyClient

client = ApifyClient("<YOUR_APIFY_TOKEN>")
run = client.actor(
    "nomad-agent/linkedin-enrich-translate-normalize-scraper"
).call(run_input={
    "schemaVersion": "nomad-agent-job-search-input-v1",
    "keyword": "data engineer",
    "location": "Germany",
    "postedWithin": "7d",
    "maxItems": 25,
}, build="latest", max_items=25, max_total_charge_usd=Decimal("0.10"))

if not run["buildNumber"].startswith("1.0."):
    raise RuntimeError("unexpected Actor build")

for item in client.dataset(run["defaultDatasetId"]).iterate_items():
    print(item["data"]["title"], item["data"]["company"]["name"])

Pricing

The Actor uses pay-per-event pricing: one base event per confirmed job plus separate events only for optional translation or successful AI enrichment. Public company-profile enrichment has no separate event. Check the Apify Pricing tab before a paid run for the current prices, and use Apify's maximum-cost-per-run setting when you need a hard budget.

Limits

• Public LinkedIn pages can change, block requests, or omit fields.
• A run returns at most 1,000 jobs, even when maxItems is 0. Large cold runs can take substantially longer and consume more compute.
• Optional enrichment improves coverage but does not make unknown source facts certain or guarantee accuracy on every future posting.
• Cross-run deduplication affects delivery, not the identity or content of a job record.

Checking optional-feature outcomes

Optional features have observable per-row outcomes. For AI enrichment, inspect llm.status and its provenance. For public company enrichment, a verified profile is a versioned object in custom, while an unavailable profile is custom: null. Minimal nomad-agent-run-summary-v4 under RUN-SUMMARY reports only the public run outcome, delivered-item count, whether results were limited, and an optional bounded retry recommendation. Source counters, diagnostics, errors, and blocking reasons remain private. The summary does not turn a missing per-row optional value into a known fact.

Privacy and removal

Contact the Actor creator through the Apify issue tab for privacy requests. For correction or source-removal requests, include the canonical public job URL or source job ID; do not post private credentials or unrelated personal data.

For lossless validation, table flattening, and integration examples, use the public integration repository.