action_batch_enrichment
ChatGPTVerify and enrich up to 50 people in ONE run, with one estimate and one approval, instead of one action per person. Each item is exactly one of: {discovery_id} (a discovered name from search_people(bucket='discovered')), {person_id} (re-verify a known contact), or {linkedin_url, company_id} (someone seen in that company's team changes, from entity_get_company_signals). Names, titles and companies are read from your own records, never typed. Duplicates collapse to one; items that cannot be resolved come back under batch.refused. Each person's company is verified again in the run and the estimate includes it. Call with confirm=false for the estimate, show it, then confirm=true with the confirmation_token. Track it with portfolio_get_job(action_id).
action_company_research_pro
ChatGPTGenerate a deep multi-source research report for a company (Pro depth — more sources, deeper analysis). Use confirm=false for cost estimate, confirm=true to execute.
action_company_research_standard
ChatGPTGenerate a multi-source research report for a company (Standard depth). Use confirm=false for cost estimate, confirm=true to execute.
action_crm_push
ChatGPTPush verified companies and their contacts to the user's connected CRM (e.g. Nutshell, Monday.com). The system auto-detects which CRM is connected — users can say 'push to CRM' or 'push to Nutshell'. Syncs company data, contacts, tags, industry, and research notes. Call with confirm=false first to see estimate (includes provider name), then confirm=true to execute.
action_discovery_enrichment
ChatGPTEnrich a discovered (UNVERIFIED) entity — either a company or person — and promote it to a fully verified record with contact channels (email, phone, LinkedIn). For PERSONS, prefer passing discovery_id (the disc_<uuid> ID returned in the unverified_discovered_persons array of entity_get_company_people). The server resolves discovery_id → person_id (creating a Person record if needed). Use person_id only when you have an already-verified Person you want to re-enrich. Use confirm=false for cost estimate, confirm=true to execute (after user approves the cost).
action_field_enrichment
ChatGPTEnrich specific missing fields on companies in a job. Use confirm=false for cost estimate, confirm=true to execute.
action_intent_rescore
ChatGPTRe-score companies in a job against an updated intent profile. Use confirm=false for cost estimate, confirm=true to execute.
action_person_enrichment
ChatGPTEnrich a person's contact details (email, phone, LinkedIn) through multi-source verification. Use confirm=false for cost estimate, confirm=true to execute.
action_quick_verify
ChatGPTVerify a single company through multi-source verification pipeline (AI, web search, LinkedIn). Provide hint fields (industry, location) to improve verification accuracy. Use confirm=false to get a cost estimate first, then confirm=true to execute. INTENT SCORING (optional but recommended): Add intent parameters to get a personalized relevance grade — same company, different purpose = different score. Three options in priority order: 1. intent_profile_id — reuse a previously created profile (zero cost) 2. intent_type — pick a preset: sales_prospect, competitor_check, partner_fit, market_intel, investment_check, or legitimacy_check. Pair with intent_context for industry. No AI cost. 3. intent_text — free-form natural language (e.g. 'Evaluating as potential acquisition target in the healthcare AI space'). Triggers one-time AI analysis; cached by text hash for repeat calls. When intent is provided, the result includes intent_score, intent_category (target/competitor/middleman/partner/acquired/out_of_scope), exclusion_flag, and intent_reasoning explaining the classification. intent_score measures FITNESS TO ACT, not confidence in the category: a confirmed competitor scores LOW.
action_research_refresh
ChatGPTRe-synthesize an existing research report with latest context (reuses cached source data — much cheaper than full regeneration). Use confirm=false for cost estimate, confirm=true to execute.
action_scan_ocr
ChatGPTExtract contact information from a business card or conference badge image. Uses AI Vision OCR for printed text and decodes any PDF417 barcode on the badge, merging both — a badge barcode often carries an email that is not printed on the badge. Use confirm=false for cost estimate, confirm=true to execute.
analysis_compare_companies
ChatGPT[CATEGORY: DEEP_ANALYSIS] Compare two companies side by side on key metrics: grade, quality, completeness, industry, status, employee count, revenue. Returns verification_sources, data_freshness_days, and source_count for each company to assess relative data reliability.
analysis_get_intent
ChatGPT[CATEGORY: DEEP_ANALYSIS] Get the intent scoring summary for a company — how well it matches the user's business intent. Returns category (target/competitor/middleman/partner/acquired/out_of_scope), intent_enhanced_alignment score (fitness to ACT, not confidence in the category — a confirmed competitor scores LOW), exclusion_flag, reasoning, adjusted_grade, research_informed status, and previous_category if reclassified after research. Use this when the user asks about fit, alignment, or classification of a company.
analysis_get_research
ChatGPT[CATEGORY: DEEP_ANALYSIS] Get the latest multi-source research report for a company including verdict, key findings, evidence_links (citation URLs), research facts with confidence scores, and intent signals. Returns error if no research exists — check entity_get_company has_research flag first.
discovery_get_capabilities
ChatGPT[CATEGORY: DISCOVERY] Get a high-level map of all InsightSignal capabilities organized by category, with recommended workflows. Call this FIRST when you are unsure which tool to use, or when the user asks a broad question. Returns tool categories (core + any admin-approved generated functions), cross-tool workflows, data quality indicators, CRM deal outcome tracking (Phase 6.3/6.4 — won/lost/win-rate), and rate/cost hints. This avoids tool sprawl by guiding you to the right 2-3 tools for any task.
entity_get_company
ChatGPT[CATEGORY: ENTITY_LOOKUP] Get full details for a specific company including latest multi-source verified data, grade (A-F), quality/completeness scores, and operational status. Returns verification_sources (which independent sources contributed), data_freshness_days, disposition, has_research (bool), has_persons (bool), person_count, and snapshot_timeline (dates + grades of all verification snapshots). When the company has a CRM integration, ALSO returns crm_outcome ('won'|'lost'|'in_pipeline'), crm_provider, crm_won_at/crm_lost_at, crm_won_source/crm_lost_source, crm_loss_reason (Phase 6.3/6.4 inbound webhooks). If CRM fields are ABSENT from the response, the company is not connected to any CRM — do NOT infer 'not won/lost' from absence; the correct interpretation is 'outside the CRM pipeline'. Use snapshot_timeline to detect data issues: if the latest snapshot has a low grade but older ones have higher grades, call temporal_get_snapshots to retrieve richer historical data. Do NOT call temporal_get_snapshots proactively — only when you notice data gaps/discrepancies in the current snapshot or the user explicitly requests history.
entity_get_company_people
ChatGPT[CATEGORY: ENTITY_LOOKUP] Get contacts associated with a specific company — BOTH verified persons AND unverified discovered persons, returned in TWO SEPARATE arrays. Response shape: {verified_persons: [...], unverified_discovered_persons: [...], verified_count, unverified_count, _data_quality_contract}. verified_persons are fully validated (readiness_grade A-F, email_deliverability, phone_valid, has_email/has_phone/has_linkedin flags) — reliable for outreach. unverified_discovered_persons are AI-surfaced names with NO contact channels yet — only person_name + title_hint (both may be approximate). They cannot be contacted until enriched. MANDATORY when listing contacts to the user: keep these two groups SEPARATE in your reply. Never present an unverified discovered person as a confirmed contact. If the user wants to reach a discovered person, call action_discovery_enrichment with the discovery_id from unverified_discovered_persons — that runs verification (LinkedIn / Hunter / phone validators) and promotes the record. Each verified person carries relationship + is_employee: an investor / board member / advisor has is_employee=false and a relationship of INVESTOR_IN / BOARD_MEMBER_OF / ADVISOR_TO. Do NOT present such a person as an employee or as someone who 'works at' the company — state their actual relationship. Only is_employee=true people are staff. Results are paginated; pagination metadata describes only the verified bucket. Check entity_get_company first to see if has_persons is true before calling this.
entity_get_company_signals
ChatGPT[CATEGORY: ENTITY_LOOKUP] Hiring, leadership and posting signals for up to 50 companies in ONE call: use it instead of calling entity_get_company once per company when you work through a list or a verification job. Select with exactly one of company_ids, domains, or job_id (the companies a verification job produced; page with next_cursor). Each row is a summary: grade, intent_fit (percent) and intent_category, disposition, red flags, CRM outcome, employees (verified band, LinkedIn count, linkedin_mismatch when they disagree), hiring (state: fetched / fetch_failed / not_fetched with the reason; open roles with their source: careers page, LinkedIn or TheirStack; joined and left in 12 months), data_quality (no_movement_12m only when the data was fetched), signals and posting_signals (each with its label and text as the platform states them, month, source and link), tech_stack, people_changes (every joined, left or title change, each marked senior and target_title_match; you decide who matters), contacts held, and the research verdict, momentum and next steps when a report exists. Report each item as observed with its month; never extrapolate. Asked ids or domains that matched nothing are in not_found; a domain matching several companies is in ambiguous with every candidate (none is picked). Use since (a month) with changed_only=true to re-read only companies with a people change or posting signal since then. For one company's full evidence (posting quotes, every role), call entity_get_company with view='detail'.
entity_get_person
ChatGPT[CATEGORY: ENTITY_LOOKUP] Get full details for a specific person including title, email, phone, LinkedIn, readiness grade (A-F), email_deliverability, email_quality_score, phone_valid, phone_line_type, phone_is_voip status, and company association. Returns has_email, has_phone, has_linkedin flags — only reference contact channels that actually exist. Includes source_count (number of independent verification sources) for trust assessment. ALSO returns the verified professional profile when available: headline, about (bio), skills (list), employment_history (list of roles with company/title/dates/is_current), education (schools/degrees), languages, and certifications. Use these to answer career-history / skills / background questions directly — do NOT say a person's history is 'not in the graph' when employment_history is populated. Empty lists mean that section hasn't been enriched yet (suggest a Sync/enrichment), not that it's unknowable.
graph_get_discovered
ChatGPT[CATEGORY: GRAPH] Get detailed discovered entity records for a company from the discovery pipeline. Returns entity_subtype (organization/product/platform/individual), is_enrichable flag, secondary_relationships (additional relationship types beyond the primary), and discovery_status (discovered/enriched/matched/dismissed). Results are paginated. Use this for detailed discovery metadata. For competitor/partner lists, prefer graph_get_relationships.
graph_get_relationships
ChatGPT[CATEGORY: GRAPH] Explore graph relationships starting from a company — branches, employees, industry connections, AND discovered entities (competitors, partners, subsidiaries, investors, customers, key people). This is the PRIMARY function for answering questions about a company's competitors, partners, or any related entities. Returns both Neo4j graph data (nodes/edges) and MongoDB discovered entities in a single response. Use this when the user asks about competitors, partners, subsidiaries, relationships, or connections.
portfolio_get_activity
ChatGPT[CATEGORY: PORTFOLIO] Get the user's recent verification runs, research reports, and enrichment jobs. Results are paginated. Use next_cursor from response to get more pages. The response carries 'as_of' (when the page was built) and 'window' {basis, since, until, unbounded}. If you pass no since_days/started_after/started_before, window.unbounded is true and 'total' is an ALL-TIME count — describe it as such, and never present those rows as belonging to a recent period. Rows may be months old; check started_at against as_of before calling anything recent.
portfolio_get_credits
ChatGPT[CATEGORY: PORTFOLIO] The workspace's credit balance as the app shows it (available_balance: what can be spent now, after holds, with any member spending cap applied; held_credits: reserved for running jobs and for this session's open read calls, settled when they finish), how many read calls one credit covers (reads_per_credit: each read call counts once, however many companies it returns), and the credits charged in the last days grouped by product (Verification, Research, Enrichment, Chat and MCP, ICP sourcing, Other). Use it before a large job instead of starting a throwaway estimate.
portfolio_get_crm_outcomes
ChatGPT[CATEGORY: PORTFOLIO] Get CRM deal outcome analytics from inbound CRM webhooks (Phase 6.3/6.4). Returns won_count, lost_count, in_pipeline_count (pushed to CRM, not yet resolved), not_in_crm_count (never pushed — NOT counted as open deals), closed_count, total_portfolio, win_rate_pct (computed only over closed deals), by_source breakdown per CRM provider (Nutshell, Monday.com, HubSpot, Pipedrive), top_loss_reasons, and sample_won / sample_lost arrays. Also returns has_crm_integration flag — when false, the user has no CRM activity yet and the result includes a 'message' field explaining the gap (do NOT report '0 wins' as deal failures). Use for questions about win rate, deal outcomes, loss reasons, or CRM performance. After calling this, drill into specifics with search_companies(crm_outcome='won'|'lost'|'in_pipeline').
portfolio_get_job
ChatGPT[CATEGORY: PORTFOLIO] One job's status and what it actually cost: status, terminal (true once completed, failed, cancelled or abandoned), waiting_on when it is paused for the user (a review, a selection, a credit top-up), progress counts, and credits_charged (what the ledger took, never the estimate). Pass job_id, or the action_id any action tool returned. An action that runs no job (contact enrichment, research, a CRM push) is answered from its own record: status, terminal, result and credits_charged, with next naming the tool that reads what it produced. Set wait_seconds (up to 60) to wait for it to finish in this one call instead of polling. When a verification finishes, next names entity_get_company_signals for its companies.
portfolio_get_stats
ChatGPT[CATEGORY: PORTFOLIO] Get big-picture statistics about the user's verified knowledge graph: average quality/completeness across all entities, grade distribution (A-F), industry breakdown, company size distribution (SMB/Mid-Market/Enterprise), revenue distribution (bucketed ranges), total companies and people counts, sample of top 5 companies (by grade and quality), and per-industry quality stats. Includes a hint for drilling down with search_companies filters. Use this when the user asks for charts, graphs, or breakdowns of their portfolio.
search_companies
ChatGPT[CATEGORY: SEARCH] Search and filter verified companies in the user's data. Returns a paginated list with grade (the EFFECTIVE InsightSignal grade — intent-adjusted when intent scoring exists, else the base data grade; each row also carries data_grade, grade_source, and numeric score), quality, completeness, source_count (number of independent verification sources), and — ONLY for records with a CRM integration — crm_outcome ('won'|'lost'|'in_pipeline'), crm_won_at, crm_lost_at, crm_won_source, crm_lost_source, crm_loss_reason. Records without any CRM connection omit the CRM fields entirely (do NOT assume absence means 'not won/lost'). Use crm_outcome='won' for closed-won deals, 'lost' for closed-lost, 'in_pipeline' for pushed-but-not-resolved, 'not_in_crm' for never-pushed records. Results are paginated. Use next_cursor from response to get more pages. Prefer filters over paging. ALL filters are optional, independent, and AND-combine — grade/min_grade work fine with no industry filter (pass industry ONLY when the user names one). Industry matching is case-insensitive and semantic; do not retry the same term with different casing.
search_count_companies
ChatGPT[CATEGORY: SEARCH] Count companies matching specific criteria. Useful for aggregate questions like 'how many tech companies do I have?' or 'how many deals did we win?'. Supports the same filter surface as search_companies.
search_events
ChatGPT[CATEGORY: SEARCH] List the events (trade shows, conferences) where this user's contacts were captured via Scan Contact, with how many records came from each. Call this BEFORE using the events filter on search_companies / search_people so you pass a real slug rather than guessing one — 'CES 2027' is stored as 'ces-2027'. Answers 'which events have I been to?' and 'how many contacts did I get at CES?'. Read-only; captures are created in the Scan Contact wizard, not from chat.
search_people
ChatGPT[CATEGORY: SEARCH] Search and filter people across the user's data — returned as TWO BUCKETS in the response: verified_persons (Person records that completed enrichment: real email/phone/LinkedIn, ready for outreach) and unverified_discovered_persons (DiscoveredEntity records — names surfaced by research/discovery, NO contact channels yet). Use the bucket parameter to scope: 'verified' for outreach questions ('who can I reach out to', 'find me deliverable emails'), 'discovered' for enrichment questions ('show me discovered names', 'who can I enrich next' — these carry discovery_id for action_discovery_enrichment), or 'both' (default) when intent is ambiguous. Verified bucket carries readiness_grade (A-F), email/phone/LinkedIn availability, enrichment_status. Discovered bucket carries person_name + title_hint + discovery_id (and next_step_to_verify hint). Cross-entity filters (industry, state) apply via knowledge graph traversal. REACHABILITY / 'who can I reach out to / contact / sell to' intent: reachability is already captured by readiness_grade (it scores email + phone + LinkedIn together), so use sort_by='readiness' and do NOT stack per-channel filters. email_deliverable / phone_verified each NARROW to that single channel and combine as AND — set one true ONLY when the user explicitly requires that specific channel; never pass false (it is not a negative filter). Verified results are paginated; discovered results are returned in full (capped at 50). Use next_cursor for more pages of verified. NEVER present a discovered person as a verified contact — they have no validated email/phone. If user wants to USE a discovered person, call action_discovery_enrichment(discovery_id=...) to promote them. Each person carries relationship + is_employee: an investor / board member / advisor has is_employee=false (relationship INVESTOR_IN / BOARD_MEMBER_OF / ADVISOR_TO) — for 'who works at' / 'employees / staff at' questions, exclude or clearly distinguish them; only is_employee=true people are staff.
temporal_get_snapshots
ChatGPT[CATEGORY: TEMPORAL] Get historical verification snapshots for a specific company — the temporal provenance trail. Each snapshot includes grade, quality, completeness, verification_sources (which tools contributed), AND field-level values with per-field provenance (csv_provided/tool_discovered/user_override), source_url (citation link), source_count, confidence, and data_freshness_days. Compare snapshots to see what changed between verifications. DUAL PURPOSE — use in two scenarios only: (1) REACTIVE: When entity_get_company shows missing data or a low grade, and snapshot_timeline reveals older snapshots with better grades — fetch those to recover richer data that a bad recent snapshot may have masked. (2) EXPLICIT: When the user asks for history, trends, or temporal comparison ('how has this changed?', 'show me snapshots', 'what was the grade last month?'). Do NOT call this proactively on every company query. Supports anchored lookup: specify around_date to get snapshots around a specific date (before AND after), useful for skipping bad snapshots. REQUIRES a company name — cannot query all companies at once.