Company search The Phase 7 capability contract owns the public schema and limits.
This page preserves reusable workflow details. It does not define a second input or result contract.
Company Search answers which companies match supported account criteria.
It returns selectable company references and observed display fields. It does not enrich, score or write CRM records.
The Companies surface carries selected refs into Company Enrich or People Search through a separate user action.
The earlier enrich: True default is superseded. V1 has no enrich, enrich_fields or depth search parameter.
Sonar remains a legacy surface during migration; it is not the new capability's product name.
company.search
-> validate source selection and bounded filters
-> read supplied identities and/or query Explorium
-> normalize identity and merge duplicate observations
-> apply deterministic gates once
-> persist tenant-owned Run output
-> return stable selectable references
ENG-2278 owns the core and supplied source. ENG-2295 owns the Explorium source.
The company.search tool uses the Explorium businesses endpoint for bounded discovery.
The company.search_supplied tool retains the free supplied-only contract and grant.
Publish a new workflow version through the standard lifecycle to use paid discovery.
The default workflow budget is 400 cents. No provider enrichment call belongs to this path.
Source V1 status Reusable detail Supplied list Supported as typed CompanyRefs Preserve input order and report unresolved rows Explorium Supported V1 provider Translate only the contract's supported firmographic filters Exa / Parallel company discovery Deferred for direct Company Search Existing planners can inform a later adapter; do not invoke signal extraction here Signals store Deferred for direct Company Search A future source reads stored signals; it must not start a new signal search
The caller names sources. No compiler silently enables another source after zero results or a provider error.
Merge provider and supplied observations before gates. Keep evidence from both within the public result limit.
Use the shared domain normalization rule. Use company LinkedIn when a valid company has no domain.
A company name is evidence, never a deduplication key. Conflicting canonical keys require an explicit conflict result.
An unresolved row stays visible in diagnostics. Do not silently drop a domainless company that has a valid LinkedIn identity.
Search returns canonical IDs only when known. It does not create an Intelligence row merely to obtain a selectable ref.
The gate applies supported filters once. It has no refill loop.
Keep a row when a field is unknown and report unknown_filter_value.
Drop a row only when an observed value contradicts a filter. Record filter_mismatch and funnel counts.
A row with unknown size does not become a verified size match.
Unsupported criteria fail validation; they do not pass through to an undisclosed trailing judge.
Deduplication, gates and result limits use stable ordering. A replay cannot choose a different winning observation.
Keep per-source status and unresolved counts when one source fails. Total source failure is not a successful empty roster.
The design scenarios cover these cases.
Signals owns signal research and scoring. It may pass discovered candidates through the supplied search source.
It then calls Company Enrich explicitly when fields are needed.
This composition preserves signal discovery without adding a web source or automatic enrichment to Company Search.
The diagram below records the superseded automatic-enrichment design.
Its enrichment step, four-source V1 claim and trailing scoring path are not the Phase 7 contract.
Company search — find the companies, and nothing else
Company search — find the companies, and nothing else Discovery picks the instrument that fits the brief. Enrichment is invoked, not embedded. A signal is read from the store, never searched again.
Discovery picks the instrument that fits the brief. Enrichment is invoked, not embedded. A signal is read from the store, never searched again. workflow/company_search.requested
workflow/company_search.requested { brief, sources[], target_count, depth, enrich } workflow.company_search · Inngest orchestrator (thin step spine)
workflow.company_search · Inngest orchestrator (thin step spine) 1 · prepareparse event · preflight · start run
1 · prepare parse event · preflight · start run 2 · compile the briefrequest → CompanySearchBrief
2 · compile the brief request → CompanySearchBrief 3 · discoverctx.group.parallel over the enabled sources
3 · discover ctx.group.parallel over the enabled sources 4 · resolve identitycandidate → one domain, then dedup
4 · resolve identity candidate → one domain, then dedup 5 · invoke enrichmentON by default — the roster comes back filled
5 · invoke enrichment ON by default — the roster comes back filled 6 · gatehard filters · one serial pass · no refill
6 · gate hard filters · one serial pass · no refill 7 · persist + finalizestore the roster, then dispatch scoring
7 · persist + finalize store the roster, then dispatch scoring Pick the instrument the brief deserves
Pick the instrument the brief deserves a firmographic brief, a thesis brief and an event brief are three different searches — the old spine ran only one of them Sources · component.company.search.source.* · each one self-gates on the brief
Sources · component.company.search.source.* · each one self-gates on the brief .explorium firmographic provider querysize · country · industry · revenue
.explorium firmographic provider query size · country · industry · revenue deterministic, no agent, one billed call .signals_store SQL read on intel_signalsthe change: a stored signal costs one query. Sonar searched the web for the same fact on every run and then threw the evidence away.
.signals_store SQL read on intel_signals the change: a stored signal costs one query. Sonar searched the web for the same fact on every run and then threw the evidence away. .exa · .parallel planner agent + web searchfor a thesis no database holds. The planner now plans a COMPANY search — the signal extractor is gone from this lane.
.exa · .parallel planner agent + web search for a thesis no database holds. The planner now plans a COMPANY search — the signal extractor is gone from this lane. .supplied the caller's own lista named list needs a gate, not a search. Import and the CRM enter here.
.supplied the caller's own list a named list needs a gate, not a search. Import and the CRM enter here. Identity resolve + dedup · deterministicone domain per candidate · dedup on the domain · a candidate that resolves to no domain is dropped
Identity resolve + dedup · deterministic one domain per candidate · dedup on the domain · a candidate that resolves to no domain is dropped it asserts identity only, and never a firmographic field workflow.enrichment · INVOKED, NOT EMBEDDED · enrich=True IS THE DEFAULT
workflow.enrichment · INVOKED, NOT EMBEDDED · enrich=True IS THE DEFAULT subject_type=company refresh=staleownership moves out, the fields do not: the result carries dict[FieldName, FilledField] — value, source and filled_at
subject_type=company refresh=stale ownership moves out, the fields do not: the result carries dict[FieldName, FilledField] — value, source and filled_at enrich=False is the opt-out, for a caller that needs identity alone Gate exclusions · country · industry · size · revenue · recencykeep on unknown. A filter drops a row only on a present, contradicting value. One serial pass has no refill, so an over-drop is permanent.
Gate exclusions · country · industry · size · revenue · recency keep on unknown. A filter drops a row only on a present, contradicting value. One serial pass has no refill, so an over-drop is permanent. what no field can prove (funding bands · SEIS/EIS · semantic excludes) does not gate — it rides to the judge Persist the ENRICHED roster — unscored workflow_run_companies
Persist the ENRICHED roster — unscored workflow_run_companies results render before the judge runs lead_score_background dispatched after finalize
lead_score_background dispatched after finalize patches badges in place What this workflow stopped doing Sonar discovered companies by searching the web for signal news, then grouped the news into companies.
What this workflow stopped doing Sonar discovered companies by searching the web for signal news, then grouped the news into companies. A firmographic brief has no news to find, so the best instrument for it never ran. It also enriched inline, so the CRM could not reach the waterfall without a fake search. Signals are a store, not a search workflow.signals_search fills intel_signals. This workflow reads it.
Signals are a store, not a search workflow.signals_search fills intel_signals. This workflow reads it. One agent run pays for the fact once. Every later brief that wants the same fact pays for one query. That is the whole reason the two were split. Why the gate stays here Only search holds the brief. Enrichment holds no opinion about a match, and it must not learn one — four other callers use it.
Why the gate stays here Only search holds the brief. Enrichment holds no opinion about a match, and it must not learn one — four other callers use it. So enrichment fills the fields and hands them back. The gate is the next step, in this workflow. Legend trigger event costs an LLM call the branch point deterministic step durable write dashed = another workflow, or a trailing lane
dashed = another workflow, or a trailing lane brief sources[] candidates domains filled fields matched Text is not SVG - cannot display Company search — find the companies, and nothing else
Company search — find the companies, and nothing else Discovery picks the instrument that fits the brief. Enrichment is invoked, not embedded. A signal is read from the store, never searched again.
Discovery picks the instrument that fits the brief. Enrichment is invoked, not embedded. A signal is read from the store, never searched again. workflow/company_search.requested
workflow/company_search.requested { brief, sources[], target_count, depth, enrich } workflow.company_search · Inngest orchestrator (thin step spine)
workflow.company_search · Inngest orchestrator (thin step spine) 1 · prepareparse event · preflight · start run
1 · prepare parse event · preflight · start run 2 · compile the briefrequest → CompanySearchBrief
2 · compile the brief request → CompanySearchBrief 3 · discoverctx.group.parallel over the enabled sources
3 · discover ctx.group.parallel over the enabled sources 4 · resolve identitycandidate → one domain, then dedup
4 · resolve identity candidate → one domain, then dedup 5 · invoke enrichmentON by default — the roster comes back filled
5 · invoke enrichment ON by default — the roster comes back filled 6 · gatehard filters · one serial pass · no refill
6 · gate hard filters · one serial pass · no refill 7 · persist + finalizestore the roster, then dispatch scoring
7 · persist + finalize store the roster, then dispatch scoring Pick the instrument the brief deserves
Pick the instrument the brief deserves a firmographic brief, a thesis brief and an event brief are three different searches — the old spine ran only one of them Sources · component.company.search.source.* · each one self-gates on the brief
Sources · component.company.search.source.* · each one self-gates on the brief .explorium firmographic provider querysize · country · industry · revenue
.explorium firmographic provider query size · country · industry · revenue deterministic, no agent, one billed call .signals_store SQL read on intel_signalsthe change: a stored signal costs one query. Sonar searched the web for the same fact on every run and then threw the evidence away.
.signals_store SQL read on intel_signals the change: a stored signal costs one query. Sonar searched the web for the same fact on every run and then threw the evidence away. .exa · .parallel planner agent + web searchfor a thesis no database holds. The planner now plans a COMPANY search — the signal extractor is gone from this lane.
.exa · .parallel planner agent + web search for a thesis no database holds. The planner now plans a COMPANY search — the signal extractor is gone from this lane. .supplied the caller's own lista named list needs a gate, not a search. Import and the CRM enter here.
.supplied the caller's own list a named list needs a gate, not a search. Import and the CRM enter here. Identity resolve + dedup · deterministicone domain per candidate · dedup on the domain · a candidate that resolves to no domain is dropped
Identity resolve + dedup · deterministic one domain per candidate · dedup on the domain · a candidate that resolves to no domain is dropped it asserts identity only, and never a firmographic field workflow.enrichment · INVOKED, NOT EMBEDDED · enrich=True IS THE DEFAULT
workflow.enrichment · INVOKED, NOT EMBEDDED · enrich=True IS THE DEFAULT subject_type=company refresh=staleownership moves out, the fields do not: the result carries dict[FieldName, FilledField] — value, source and filled_at
subject_type=company refresh=stale ownership moves out, the fields do not: the result carries dict[FieldName, FilledField] — value, source and filled_at enrich=False is the opt-out, for a caller that needs identity alone Gate exclusions · country · industry · size · revenue · recencykeep on unknown. A filter drops a row only on a present, contradicting value. One serial pass has no refill, so an over-drop is permanent.
Gate exclusions · country · industry · size · revenue · recency keep on unknown. A filter drops a row only on a present, contradicting value. One serial pass has no refill, so an over-drop is permanent. what no field can prove (funding bands · SEIS/EIS · semantic excludes) does not gate — it rides to the judge Persist the ENRICHED roster — unscored workflow_run_companies
Persist the ENRICHED roster — unscored workflow_run_companies results render before the judge runs lead_score_background dispatched after finalize
lead_score_background dispatched after finalize patches badges in place What this workflow stopped doing Sonar discovered companies by searching the web for signal news, then grouped the news into companies.
What this workflow stopped doing Sonar discovered companies by searching the web for signal news, then grouped the news into companies. A firmographic brief has no news to find, so the best instrument for it never ran. It also enriched inline, so the CRM could not reach the waterfall without a fake search. Signals are a store, not a search workflow.signals_search fills intel_signals. This workflow reads it.
Signals are a store, not a search workflow.signals_search fills intel_signals. This workflow reads it. One agent run pays for the fact once. Every later brief that wants the same fact pays for one query. That is the whole reason the two were split. Why the gate stays here Only search holds the brief. Enrichment holds no opinion about a match, and it must not learn one — four other callers use it.
Why the gate stays here Only search holds the brief. Enrichment holds no opinion about a match, and it must not learn one — four other callers use it. So enrichment fills the fields and hands them back. The gate is the next step, in this workflow. Legend trigger event costs an LLM call the branch point deterministic step durable write dashed = another workflow, or a trailing lane
dashed = another workflow, or a trailing lane brief sources[] candidates domains filled fields matched Text is not SVG - cannot display Legacy reference only. Phase 7 removes automatic enrichment and limits direct search to supplied identities and Explorium. 