Entity People Search

entity-people-search

Direct lookup of the canonical LinkedIn people entity via ask_onfire. The people table ONFIRE.PEOPLE is the contact entity in the semantic model. ask_onfire does not take SQL — it takes a structured QueryIR (query={...}) that the server compiles for you. No external API hop.

When to use this

• "Find security engineers at Meridian Bank"
• "Who are the CISOs in financial services in Germany?"
• "Show me people whose job summary mentions Sentinex"
• "Look up this LinkedIn URL: linkedin.com/in/johndoe"
• "Get me VPs of Engineering at companies with 200-500 employees"
• Any people lookup that can be expressed as filters on LinkedIn profile data

Skip this for:

• Prospecting with Forschung's persona/event/technology scoring → use ai_prospecting
• Looking up a single person by name only (no company context) → use match_person first
• People who joined specific communities → use the community-join-signals skill

ask_onfire shape

ask_onfire(query={
  entity: "contact",
  select: ["<field>", ...],          // attributes / dimensions on the entity
  filters: [{dimension, op, value}], // op: eq | in | gte | lte | contains
  insight_filters: [{kind, value}],  // persona / technology concepts (resolved server-side)
  joins: [{entity, filters, insight_filters}],  // filter-only single hop; cannot SELECT joined columns
  distinct_by: "<field>",
  order_by: [{field, direction}],
  limit: <int>                       // ROW BUDGET — see Billing
})

Fields

Selectable attributes (returnable, NOT filterable)

Filterable dimensions (use in filters)

DELETED_AT is handled for you — the contact entity is filtered to live profiles server-side. You no longer write a delete filter.

Persona / seniority / technology → insight_filters (NOT dimension filters)

Role, department, seniority, and technology/vendor are not columns — they live in the insight catalog and go in insight_filters as concepts (one concept per filter; multiple insight_filters are AND-ed):

• {kind: "persona", value: "<role/department>"} — e.g. "appsec", "CISO", "data engineering".
• {kind: "persona", value: "seniority_executive"} — seniority is a persona, not a dimension. Use seniority_executive (covers VP / C-suite) / seniority_director / seniority_teamlead / seniority_ic.
• {kind: "technology", value: "<tool/vendor>"} — e.g. "Sentinex". This is the correct replacement for the old "keyword in job summary" lookup when the keyword is a technology or vendor.

Values are resolved server-side to an exact insight_name; if unsure, confirm the canonical value with resolve_insights (carrying kind) first. The server bounces back suggestions if it can't resolve.

Billing — read before calling

ask_onfire bills the client 1 credit per row returned.

• Always set a small explicit limit (the row budget). Set it to exactly the number the user asked for; for a quick look 5–10 rows is plenty.
• If you do not know how many rows the user wants, leave limit unset: ask_onfire then returns how many rows match without billing so you can confirm the count and resubmit.
• If the response is needs_confirmation (the row-budget gate — triggered when limit is unset or larger than ~50 rows and not yet confirmed), nothing was billed and no rows came back. Lower limit to what you actually need and resubmit. Do not reflexively set confirmed: true just to push a large pull through.

QueryIR templates

By company LinkedIn URL

ask_onfire(query={
  entity: "contact",
  select: ["linkedin_url", "full_name", "job_title", "seniority_levels",
           "job_summary", "location_name", "job_start_date"],
  filters: [{dimension: "current_company_url", op: "eq",
             value: "linkedin.com/company/meridian-bank"}],
  order_by: [{field: "job_start_date", direction: "desc"}],
  limit: 50
})

By job title keyword

ask_onfire(query={
  entity: "contact",
  select: ["linkedin_url", "full_name", "job_title", "seniority_levels",
           "current_company_name", "current_company_url"],
  filters: [
    {dimension: "job_title", op: "contains", value: "security engineer"},
    {dimension: "location_country", op: "eq", value: "united states"}
  ],
  limit: 50
})

By technology footprint (replaces "keyword in job summary")

A tool/vendor keyword (e.g. Sentinex) is a technology insight, not a profile substring — pass it as an insight_filter, not a job_summary match (job_summary is selectable but not filterable):

ask_onfire(query={
  entity: "contact",
  select: ["linkedin_url", "full_name", "job_title",
           "current_company_url", "current_company_name", "job_summary"],
  filters: [{dimension: "current_company_url", op: "eq",
             value: "linkedin.com/company/northwind"}],
  insight_filters: [{kind: "technology", value: "Sentinex"}],
  limit: 30
})

insight_filters AND together — to cover several technologies, run one query per technology and merge (there is no OR).

By seniority (a persona, not a column)

Seniority is not a dimension — filter it with a persona insight_filter (VP / C-suite map to seniority_executive):

ask_onfire(query={
  entity: "contact",
  select: ["linkedin_url", "full_name", "job_title", "seniority_levels",
           "current_company_name", "current_company_industry", "location_country"],
  filters: [{dimension: "current_company_industry", op: "contains", value: "financial"}],
  insight_filters: [{kind: "persona", value: "seniority_executive"}],
  limit: 50
})

By LinkedIn URL (single person lookup)

ask_onfire(query={
  entity: "contact",
  select: ["linkedin_url", "full_name", "job_title", "seniority_levels",
           "current_company_name", "current_company_url", "location_name",
           "location_country", "summary"],
  filters: [{dimension: "linkedin_url", op: "eq",
             value: "https://www.linkedin.com/in/johndoe"}],
  limit: 1
})

Output handling

ask_onfire returns a dataset handle + preview_rows (first 20).

1. Show preview_rows to the user as a formatted table.
2. State total_count — if it's more than 20, tell the user the full set is in the dataset.
3. Offer download_dataset(dataset_id="...") for the full CSV.
4. For follow-up filters ("show me only the VPs", "add their company size"), use query_datasets against the dataset_id — do not re-run ask_onfire (re-running rebills per row).

What this entity CANNOT express

ask_onfire is a structured query, not SQL. The following are not expressible against contact — do not try to fake them:

• Filtering on skills — skills is an array dimension, not scalar-filterable. If the user wants people by a skill tag, that filter cannot be applied here; the closest expressible proxy is a technology insight_filter when the skill is actually a tool/vendor.
• Arbitrary free-text matches on job_summary / summary / headline — these are selectable attributes, not filterable. If the keyword is a technology/vendor → use insight_filters kind=technology (above). If it is a role/department → use insight_filters kind=persona. If it is genuinely arbitrary free text with no insight equivalent, it cannot be filtered here — say so rather than fabricating a filter.
• GROUP BY distributions, aggregations beyond contact_count, window functions, or temporal math — not supported.

Common pitfalls

• Putting persona / role / seniority / technology in filters — they are not columns. They go in insight_filters (seniority as kind: "persona", value seniority_executive etc.).
• Trying to filter job_summary/summary/headline — selectable only. For a tool/vendor keyword use a technology insight_filter instead.
• Leaving limit unset on purpose — that triggers the row-budget gate and returns no rows. Set the budget to what the user actually wants.
• Re-running ask_onfire for "show me only the engineers" — slice the existing dataset with query_datasets to avoid rebilling.