add_contact_note
ChatGPTAdd a freeform timestamped note to a contact. Use this for 'leave a note', 'log a call', 'record a meeting'. Use update_contact for editing profile fields like email or job title — not this.
add_contact_tag
ChatGPTAttach a tag to a contact, looked up by tag name. Idempotent. Creates the tag if it doesn't exist yet (with default gray color).
archive_contact
ChatGPTSoft-delete a contact: sets isArchived=true. Recoverable from the web UI. Use when the user says 'remove', 'delete', or 'archive' a contact — hard delete is web-only by design.
archive_task
ChatGPTArchive a task (soft delete — hidden from the active list but recoverable from the web UI). Pass the taskId from list_tasks.
cancel_catch_up
ChatGPTCancel a catch-up that is being drafted (ACCEPTED) or in progress (ACTIVE). Stops the sequence with a manual stop: nothing further sends. Emails already sent are not recalled, and a finished catch-up can't be canceled. This is not undoable from the connector.
cancel_introduction_followup
ChatGPTCancel a SCHEDULED post-intro follow-up email (network or client side) so the hourly sender will not send it — e.g. when the two parties already connected so the 'did you connect?' check-in is no longer needed. Only works before the follow-up has been sent; an already-sent follow-up cannot be canceled. Reversible with restore_introduction_followup. To change WHEN it sends instead of stopping it, use update_introduction_draft with networkFollowUpDays / clientFollowUpDays.
cancel_team_request
ChatGPTWithdraw an intro request you sent to a teammate while it's still pending. Pass the request id from list_team_requests. Only call when the user asks to cancel.
complete_task
ChatGPTMark a task complete (or reopen it by passing completed=false). Pass the taskId from list_tasks.
create_company
ChatGPTAdd a company the advisor ALREADY has a working relationship with (they advise it, invested in it, sit on its board, or lead it) and has said so. Never add a prospect, a company still negotiating an advisory arrangement, or a company merely mentioned in mail or a meeting: for those, add a contact note or mention it in your summary instead. name and primaryContact are required — every company needs at least one person attached, or intro drafting downstream has no one to reach. For primaryContact, pass existingContactId (resolve with search_contacts / query_contacts) OR firstName + lastName + email; a known email is matched to an existing contact instead of duplicating them. website is recommended (the domain is extracted from it and powers intro matching). Optionally set description, industry, stage, roles, currentNeeds, needsDescription. Duplicate names are rejected — the existing companyId is returned. Note: the relationship snapshot and AI health analysis run in the web app, not here.
create_contact
ChatGPTManually create a new contact in the advisor's network. Requires at least one of (fullName, firstName/lastName) and at least one identifier (email or linkedinUrl). source is set to MANUAL. Duplicate emails are rejected — the existing contactId is returned instead.
create_introduction_draft
ChatGPTUse ONLY when the user wants to START or SEND a real introduction (opt-in emails / sending). If they just want a message or follow-up written to copy and send themselves, do NOT call this — draft it directly in chat. Without drafts this is a heavy call: it generates the emails via an LLM and opens a confirmation iframe. Create a DRAFT introduction. Supports two shapes: - contact → portfolio company: pass contactId + companyId. - contact ↔ contact (peer intro): pass contactId + clientContactId. Role semantics (READ CAREFULLY — getting this backwards swaps the roles in every generated email) - contactId = NETWORK side. The contact being introduced — i.e. the person from the advisor's broader network whose name/credentials are being sent over. Think "the asset the advisor is offering." In the generated emails this person's job title, company, and LinkedIn are pitched TO the other side. - clientContactId = CLIENT side. The intro recipient — i.e. the person the advisor advocates for (typically a founder / portfolio-company operator the advisor works with). The network contact is being presented to them. - companyId = CLIENT side, company shape. Same role as clientContactId — the portfolio company receiving the intro — but resolves the recipient from the company's point-of-contact. Mnemonic for "introduce A to B" phrasing: A is being introduced (sent over) → A = contactId. B is receiving the introduction → B = clientContactId (or companyId). - "introduce Joel to Bob" → contactId = Joel, clientContactId = Bob. - "connect Joel with Bob, my founder" → "my founder" tags Bob as the client side (the recipient) → contactId = Joel, clientContactId = Bob. - "intro Bob (my portco CEO) to Joel" → Bob is the client / recipient, Joel is the network contact being sent his way → contactId = Joel, clientContactId = Bob. (Don't be fooled by sentence order; the portfolio party is always the client side.) If the user's phrasing genuinely doesn't disambiguate which person is the network contact vs. the client recipient (e.g. neither is obviously a portfolio/founder and neither was previously discussed in that role), ASK before calling — do not 50/50 guess. A wrong assignment is wasted because the generated emails frame the wrong person as the recipient. Resolution FIRST resolve real ids: use search_contacts/query_contacts for the contact and list_companies/search_companies for the company, and make sure you have THE specific person/company the user means. If a name is ambiguous (multiple matches), list the matches and ask the user which one they mean — do NOT guess. (The search results card is informational; the user answers in chat.) This generates 5 emails via an LLM call, so don't fire it on a half-resolved name. Exactly one of companyId or clientContactId is required. By default the server runs the same AI email generator the web wizard uses and persists the drafts (network opt-in, client opt-in, intro, two follow-ups, two no-reply reminders) on the introduction record. Renders ui://intro-confirm — the iframe is the user's confirmation surface to choose a consent model, regenerate with feedback, edit, and Send. Your own wording (drafts) When the user has already written or approved the emails in this chat, pass them as drafts instead of letting the server generate and then replacing them with update_introduction_draft (that leaves a first version the user never wanted and makes the card and chat disagree). Keys match update_introduction_draft's emailType: networkOptIn, clientOptIn, intro, networkFollowUp, clientFollowUp, networkApprovalReminder, clientApprovalReminder; each {subject, body}. Supplied text is used verbatim (the advisor's signature is appended if it isn't already there). If you supply every email the consent model sends first (DOUBLE_OPT_IN: networkOptIn + clientOptIn + intro; NETWORK_FIRST_SINGLE_OPT_IN: networkOptIn + intro; CLIENT_FIRST_SINGLE_OPT_IN: clientOptIn + intro; DIRECT_INTRO: intro), nothing is generated: it is fast, and any follow-up or reminder you left out gets a standard template the user can …
create_tag
ChatGPTCreate a new tag (or update an existing tag's color/description). Tags are unique per advisor by name. For just attaching a tag to a contact use add_contact_tag — it upserts the tag automatically.
create_task
ChatGPTCreate a task (to-do / follow-up). Pass either text (natural language like 'call Sarah about funding by Friday, high priority' — the server extracts title, due date, and priority) OR structured fields (title required). Optionally link the task to contacts and/or companies by id (resolve them first with search_contacts / search_companies). If both text and title are given, the structured fields win.
delete_contact_note
ChatGPTPermanently delete a contact note. Irreversible — only use when the user explicitly asks to delete a specific note (by noteId).
dismiss_catch_up
ChatGPTDismiss a SUGGESTED catch-up so it stops surfacing. A plain dismiss pauses suggestions for that contact for 60 days; neverSuggest: true opts the contact out of catch-up suggestions permanently. Pass reason when the user says why: NOT_A_RELATIONSHIP (vendor, past client) or NO_DIRECT_CONTACT (client reached through an agency or partner) are permanent and tag the contact "No catch-ups"; NEVER_REPLIES waits a year; BAD_TIMING (on leave, busy) snoozes 90 days; NEVER_SUGGEST means this person must never be suggested anywhere again (archives the contact, record kept; their suggested intros are dismissed and any catch-up sequence in flight stops); OTHER is the plain 60 days. Only a suggestion can be dismissed; for one that is already drafting or sending, use cancel_catch_up.
dismiss_introduction
ChatGPTDismiss a suggested introduction (sets its status to DISMISSED) so it no longer surfaces as a recommendation. Used by the recommendation card's Dismiss action. Only affects intros that haven't been started; once an intro is in flight, use mark_introduction_outcome instead. When the user says one of the people should never be suggested again (anywhere: catch-ups, introductions, search), pass that person's id as neverSuggestContactId: the contact is archived (record kept), their pending catch-up suggestion and suggested intros are dismissed and any catch-up sequence in flight stops.
draft_catch_up
ChatGPTStart a catch-up with one contact: an opening email plus optional reminders that send in the same thread and stop the moment anything happens between the two of them. Renders ui://catch-up-confirm, the card where the user reviews every step, edits it, and sends. Two ways to supply the emails: 1. PASS YOUR OWN (steps) — preferred whenever you have already written the wording in chat and the user liked it. Send exactly those emails: steps[0] is the first email ({subject, body}), each later entry is a reminder ({body, optional offsetDays}). Nothing is regenerated, so the user gets the text they just approved, and it is fast (no generation call). Reminders have no subject of their own; they send as "Re: " plus the first email's subject. 2. OMIT steps — the server writes the whole sequence itself in one generation call. Use this when the user wants NetworkOS to draft it and you have no approved wording. Takes a moment. Reserve this tool for actually starting a sequence. If the user only wants wording to read or copy, write it in your reply instead. Resolve the contactId FIRST with search_contacts or query_contacts. If an in-flight catch-up already exists for the contact, steps overwrite its drafts, while omitting steps returns the existing drafts unchanged (no duplicate row, no rewrite); an already-sending one returns its status instead. If an old finished or dismissed catch-up exists, it is recycled into a fresh one: explicit user intent overrides every cooldown, including 'don't suggest again'. The contact must belong to the advisor, be active, and have an email address. NOTHING sends from this tool. The user reviews on the card, or you can edit with update_catch_up and rewrite with regenerate_catch_up_drafts; only send_catch_up (on an explicit instruction to send) sends the first email.
enrich_contact
ChatGPTRun data enrichment on a contact to fill missing profile fields — including a missing email address. For a missing email it searches the advisor's own connected Gmail/Calendar correspondence FIRST (a verified address seen in real threads) and only falls back to external vendor data when the mailbox has nothing. A vendor-sourced address is saved as unverified and may be replaced by a later mailbox-confirmed match; addresses the advisor curated are never overwritten. Other profile fields are only filled when currently empty.
find_contact_email
ChatGPTSearch the advisor's connected Gmail inboxes AND Google Calendar events for a contact's likely email address, matched by name against message headers and event attendees. Use when a contact has no email on file (e.g. an introduction is blocked). Returns up to 3 ranked candidates with occurrence counts — confirm with the user, then save the right one via update_contact.
find_contact_linkedin
ChatGPTLook up ONE contact's LinkedIn profile URL from their email address through the paid enrichment vendor and save it to the contact. The introduction card shows a button for this when a party has no profile on file, so the advisor decides when to spend the credit. Only call it yourself when the user explicitly asks for this specific person's LinkedIn profile to be found. NEVER call it to backfill profiles across several contacts or without being asked: every call costs enrichment credits. A miss is a normal result (found: false), not an error.
find_intro_opportunities
ChatGPTCompute introduction opportunities between the advisor's network and a specific company at request time. Builds a semantic search from the company's name, industry, currentNeeds (tags), and needsDescription (free text), ranks candidates by relevance + relationship strength, and filters out anyone already in flight (or terminal) for that company. Use when the user asks 'who in my network should I introduce to [company]?' or any variant. Pass companyId when known, otherwise companyName. Renders a multi-select picker (ui://contact-picker): the user picks one or more candidates and hits Next, which asks you to create an introduction draft for each selection. If your final recommendation is a curated subset drawn from several lookups instead of this tool's raw output, present it with propose_introductions. Tip: results improve markedly when the company has currentNeeds or needsDescription set — surface that nudge to the user if the empty-state message says needs are missing.
get_catch_up
ChatGPTFetch one catch-up by id with everything on it: contact, status, the reason it was suggested, every step's draft with sent/scheduled state, the reminder offsets (each reminder sends a set number of days after the previous email's actual send), Gmail thread and reply state, stop reason, and outcome timestamps. Renders ui://catch-up-confirm so the user can review, edit and send it inline. Use before editing, regenerating, or sending, or to answer 'where does my catch-up with X stand?'. A CANCELED catch-up with resumable: true can be resumed from its page in the web app.
get_catch_up_touchpoints
ChatGPTThe history between the advisor and a catch-up's contact: every recorded touchpoint in the last 3 years (calendar meetings, emails sent, replies received, notes) with dates, oldest first, plus a count by kind. Read-only and best-effort: calendar meetings need a connected Google account and may be missing when one mailbox is unreachable. The ui://catch-up-confirm card calls this itself to draw its history strip, so you rarely need to; use it to answer 'when did we last meet?' or 'how often have we talked?'.
get_contact_detail
ChatGPTShow a rich inline detail card for one contact: profile, recent notes, tags, and in-flight introductions, with Add note / Draft intro / Open actions. Use when the user says 'show me [person]', 'open [person]', 'tell me about [person]', or taps a contact. Resolve the contactId with search_contacts or query_contacts first. Renders ui://contact-detail.
get_introduction
ChatGPTFetch a single introduction by id with all persisted email drafts (the 5 core drafts plus the two consent nudges networkApprovalReminder / clientApprovalReminder), status, full timeline, contact/company/clientContact info, any received replies, followUps: each person's post-intro follow-up with its timing in days, when it is due (sendsAt) or whether it was sent/canceled, recipients (To + the intro's cc/bcc) and draft, and reminders: each opt-in's no-reply reminder with its timing in days (networkApprovalReminderDays / clientApprovalReminderDays), due date and whether it sent. Use it to answer 'when does the follow-up go out?' or 'when does the reminder go out?'. Use this to inspect or remind yourself of intro context before regenerating, sending, or marking outcome.
get_network_insights
ChatGPTGet a high-level summary of the advisor's network: total contacts, total companies, intros in flight, notes added recently. Renders ui://network-insights as an inline dashboard card.
get_sync_status
ChatGPTCheck the contact-sync pipeline: whether a sync is running, overall progress, per-phase job status (Google Contacts / Gmail / Calendar → dedup → enrichment → classification), and the current usable contact count. Use for 'is my sync done?' or before/after trigger_sync.
get_target_list
ChatGPTEXPERIMENT (allowlisted advisors only). Re-open a company's target list: who on it is already in the advisor's network, as a multi-select picker (latest list unless listId is given). Re-runs the match against the network first, so people added since the upload count.
list_catch_ups
ChatGPTList the advisor's catch-ups: suggested reconnects with people they have real history with but haven't talked to in a while, plus sequences being drafted, in progress, and finished. Filter with view: suggested (awaiting a decision), active (drafting or sending), done (finished, includes dismissed), or all (default: everything except dismissed, in-flight rows first). query matches the contact's name or company. Each row carries the contact, status, the one-line reason it was suggested, when they last talked, and the next step. A CANCELED row with resumable: true was stopped automatically (a new touchpoint or a paused mailbox) and can be resumed from the catch-up's page in the web app. Read-only and fast.
list_companies
ChatGPTList the advisor's portfolio companies (no query required). Use for 'what companies do I support?' or to seed an intro flow. search_companies is still better when the user names a company. Includes current needs and AI-derived relationship health.
list_contact_notes
ChatGPTList timestamped notes for a contact (most recent first). Use to recall past meetings, calls, and freeform notes you've left on a person. Paginated.
list_introductions
ChatGPTList the advisor's introductions, most recent first. Optionally filter by status, contactId, or companyId. The status values are enumerated on the parameter itself, so use one of those exactly. Each row carries sequence, one line on where its email sequence stands ('Introduction sent 2026-10-05; next: follow-up to Clay on 2026-10-26'), plus lastEmailSent and nextEmail as objects; answer 'did the follow-up go out?' / 'what goes out next?' from these without fetching each intro.
list_tags
ChatGPTList all tags in the advisor's account with usage counts. Use before adding a tag to a contact so you can pick an existing tag name or notice when a new one should be created.
list_target_lists
ChatGPTEXPERIMENT (allowlisted advisors only). The target lists saved on a company, newest first, with how many of each list's people are in the advisor's network.
list_tasks
ChatGPTList the advisor's tasks (to-dos / follow-ups), soonest due first. Defaults to active (not completed, not archived). Filter by status, by a linked contactId or companyId, or by priority. Use this for 'what's on my list?', 'what's due?', or 'what follow-ups do I have for Acme?'.
list_team_requests
ChatGPTList the introduction requests you've sent to team members and their status: pending (not sent yet, even if the owner has started a draft), accepted (the owner sent the first email), declined, cancelled, or completed (the introduction itself went out). Use it to check on a request or find the id of one to cancel.
log_touchpoint
ChatGPTRecord that the advisor reached out to or spoke with a contact outside NetworkOS email (text, call, in person, LinkedIn). Call it only when the user says it happened, never on inference. Counts toward warmth and closes that contact's open catch-up as done; nothing is emailed. Pass catchUpId only to name the exact catch-up to close. requestId is only for retrying the exact same entry after an error; never reuse one for a different entry.
mark_catch_up_replied
ChatGPTMark a running catch-up sequence as replied when the contact wrote back and NetworkOS cannot see that inbox. Stops the remaining reminders and records the reply for warmth.
mark_introduction_outcome
ChatGPTClose the loop on an introduction by recording a terminal outcome and (optionally) freeform outcome notes. Use this when the user says 'mark X as completed' / 'they didn't respond' / 'they declined'. Sets the intro's status to the chosen outcome and stamps outcomeAt.
pause_introduction
ChatGPTPause an introduction's automatic emails (consent nudges, post-intro follow-ups, and the advisor's day-before heads-up) without declining it, e.g. when one person is on leave. The intro keeps its status. On the chosen date the advisor gets one reminder email; nothing resumes on its own until resume_introduction (or the email's Resume link). Use this instead of marking an intro rejected just to stop its follow-ups. Fails if the intro is already paused (pass extend: true to move the date) or has nothing scheduled.
propose_introductions
ChatGPTPresent your FINAL shortlist of recommended introductions after researching with search_contacts / query_contacts / find_intro_opportunities. Renders a multi-select picker (ui://contact-picker): the user picks one or more contacts and hits Next, which sends you a message asking to create an introduction draft for each selection — treat that message as explicit intent to start the intros. Do NOT call this for interim lookups — those already render informational result cards. Call it once, at the end, with the curated list you actually recommend (1-20 contacts, with a one-line reason each). Target semantics mirror create_introduction_draft: contactIds are the NETWORK side (the people being introduced / sent over). The optional target is the CLIENT side — the intro recipient — pass companyId for a portfolio company or clientContactId for a peer intro recipient. Omit the target only if it's genuinely not settled yet; the picker will then ask you to confirm it after selection.
query_contacts
ChatGPTSearch the advisor's contacts directly by writing a SQL WHERE clause against the contacts table. You — Claude — write the query; the server validates and executes it. Faster and more flexible than a natural-language search. Use this instead of search_contacts for any non-trivial search (semantic, attribute-based, or compound). Renders an informational results card (ui://contact-picker) showing what came back — the user is not expected to act on it. Columns you can filter on IDENTIFIERS: id (the contact's id — the same value other tools return/accept as contactId; e.g. id = 'abc123') TEXT (use ILIKE '%term%' for case-insensitive substring matching): full_name, first_name, last_name, email phone ← sparsely populated (CSV/LinkedIn imports and manually-added contacts; Google sync does not import numbers). phone IS NOT NULL answers "who can I call or text". The number itself is returned in the results. current_company, company_description, domain, industry job_title, headline, bio, description location notes ← RICHEST field. ~75% of contacts have it. Free text containing LinkedIn data, org memberships, networks, education, expertise, portfolio. Search it heavily for semantic queries. contact_type EXACT values only: 'INVESTOR', 'B2B_EXEC', 'TALENT', 'PENDING_REVIEW', 'OTHER'. A value outside this set is rejected. NUMERIC / DATES: relationship_score (integer 0-100; higher = stronger relationship; >=70 ~ warm) meeting_count, email_thread_count, interaction_count last_interaction (timestamp) created_at (when contact was added) ARRAYS: skills (text[]; use: 'react' = ANY(skills)) sources ContactSource[] (an ENUM array, not text). EXACT values only: 'GOOGLE_CONTACTS', 'GOOGLE_OTHER_CONTACTS', 'GMAIL', 'CALENDAR', 'LINKEDIN_CSV', 'LINKEDIN_EXTENSION', 'MANUAL', 'ENRICHMENT', 'DIRECTORY_SCRAPE', 'TEAM', 'CLIENT_PORTAL'. A value outside this set aborts the query. Filter with ANY, never =: 'LINKEDIN_CSV' = ANY(sources) LINKEDIN (bare) is NOT a value. LinkedIn contacts are LINKEDIN_CSV (connections export) or LINKEDIN_EXTENSION (Chrome extension), so "LinkedIn contacts" usually means both: ('LINKEDIN_CSV' = ANY(sources) OR 'LINKEDIN_EXTENSION' = ANY(sources)) Operators ILIKE '%term%' case-insensitive substring AND, OR, NOT, () IS NULL / IS NOT NULL IN ('a','b'), BETWEEN x AND y >, <, >=, <=, =, != ANY(array_col) Forbidden: semicolons, comments (-- or / /), UNION/EXCEPT/INTERSECT, subqueries selecting from non-allowed tables, any DDL/DML, set_config, pg_sleep, etc. (Server enforces.) Patterns for common asks "climate hardware founders": (job_title ILIKE '%founder%' OR job_title ILIKE '%CEO%' OR headline ILIKE '%founder%') AND (company_description ILIKE '%climate%' OR notes ILIKE '%climate%' OR industry ILIKE '%climate%') AND (company_description ILIKE '%hardware%' OR notes ILIKE '%hardware%' OR industry ILIKE '%hardware%') "ex-Stripe operators": (notes ILIKE '%Stripe%' OR headline ILIKE '%Stripe%' OR bio ILIKE '%Stripe%') AND current_company NOT ILIKE '%Stripe%' "people I've met with at least 3 times": meeting_count >= 3 "contacts in SF or NYC": location ILIKE '%San Francisco%' OR location ILIKE '%SF%' OR location ILIKE '%Bay Area%' OR location ILIKE '%New York%' OR location ILIKE '%NYC%' Tips - Column names are snake_case (full_name, contact_id) — camelCase variants like contactId are auto-normalized, but prefer snake_case. - Cast a wide net with OR'd ILIKEs across multiple text fields. Better to over-match than miss someone. - For 'notes' the data varies — search same-concept terms in headline + notes + bio + company_description. - Do NOT include advisor_id — it's auto-injected and the query is scoped to your network. - Do NOT include the WHERE keyword — just the body of the clause. - orderBy defaults to relationship_score DESC. Other allowed columns: relationship_score, meeting_count, email_thread_count, interaction_count, last_interaction, email_last_date, created_at, full_name, current_company, job_title. If validation rejects your clause, the error tells you what's wrong — adjust and retry.
recommend_introduction
ChatGPTRecord a recommended introduction so it surfaces on the advisor's dashboard for them to act on. This is the 'superconnector loop' write-back: after you've reasoned about who would benefit from meeting whom, call this to save the suggestion. It does NOT draft or send any email — the advisor reviews the recommendation and starts the double opt-in themselves. Two shapes: • Peer intro — two people in the network: pass networkContactId + targetContactId. • Advisor→company intro — a network contact who fits a portfolio company: pass networkContactId + companyId. Provide exactly one of targetContactId or companyId. Always include a concrete reason explaining the mutual benefit. This call is idempotent per pair: recommending the same pair again refreshes the existing suggestion instead of creating a duplicate, so it is safe to run on a daily cadence.
regenerate_catch_up_drafts
ChatGPTRewrite a catch-up's drafts with fresh generated text, optionally steered by instructions ('shorter', 'mention the fundraise'). Rewrites every step by default and OVERWRITES the current unsent drafts, including the user's manual edits; pass stepIndex to rewrite only that one step and keep the others verbatim. stepCount (1 to 3) changes how many steps the rewritten sequence has. Takes a moment: one server-side generation call. Only works before the first email sends. Returns refreshed structuredContent for the ui://catch-up-confirm card. For a small wording tweak the user has already dictated, prefer update_catch_up (no generation); to replace the drafts with wording YOU wrote, use update_catch_up too, which needs no generation call either.
regenerate_introduction_drafts
ChatGPTRe-run the AI email generator for an EXISTING intro draft. Heavy call (regenerates emails via an LLM) — only use when an intro draft already exists and the user wants to revise it; never for drafting a plain message in chat. Use when the user asks to change the consent model (DOUBLE_OPT_IN / NETWORK_FIRST_SINGLE / CLIENT_FIRST_SINGLE / DIRECT_INTRO), to swap the order / direction of the intro (swapContacts), or to apply new instructions like 'make it shorter' or 'mention our shared time at Stripe'. Instructions REVISE the drafts as they stand (including copy the user wrote in Claude and edits made on the card), leaving untouched what they don't mention; a consent-model change or a swap regenerates from scratch. Returns refreshed structuredContent for the ui://intro-confirm iframe.
remove_contact_tag
ChatGPTDetach a tag from a contact. Removes the ContactTag association only — the tag itself (and its other contacts) are unaffected.
request_team_intro
ChatGPTAsk a TEAMMATE for a warm introduction to one of their contacts (surfaced by team_search). The relationship owner gets an email and can accept or decline — nothing is sent to the contact yet, and you can cancel while it's pending. Pass the contactId from a team_search result. Only call this when the user has explicitly asked to request the intro. Track or withdraw requests with list_team_requests / cancel_team_request.
restore_introduction_followup
ChatGPTRestore a previously canceled post-intro follow-up email (network or client side) so it will be sent on schedule again. Only works before the follow-up has been sent.
resume_introduction
ChatGPTResume a paused introduction's automatic emails. Their timers restart from now (a 21-day follow-up goes out 21 days after the resume), so nothing sends right away. Fails if the intro is not paused.
search_companies
ChatGPTSearch portfolio companies the advisor supports by name. Use when the user mentions a company.
search_contacts
ChatGPTSearch contacts by name, email, or company. Use when the user mentions a person by name or wants to find someone in their network. Renders an informational results card (ui://contact-picker) showing what came back — the user is not expected to act on it. For attribute/semantic/compound searches, prefer query_contacts.
send_catch_up
ChatGPTSend step 1 of a catch-up now, from the advisor's own Gmail. NEVER call this without an explicit user instruction to send this specific catch-up; drafting or reviewing is not consent to send. Exactly what happens: step 1 emails the contact immediately from the advisor's Gmail, and any remaining reminder steps then send automatically in the same thread on the schedule shown on the catch-up (each a set number of days after the previous email, during working hours). The automatic reminders stop on their own the moment anything happens between the advisor and the contact: a reply, a meeting, or any other recorded touchpoint. Only works on a reviewed draft (status ACCEPTED) whose step 1 has not sent. Practice (demo) catch-ups never send.
send_introduction_email
ChatGPTSend one of the persisted intro emails (network opt-in, client opt-in, or final intro) via the advisor's Gmail. Called by the ui://intro-confirm iframe when the user clicks Send — usually not invoked directly by the model. Picks the right status transition based on emailType.
submit_feedback
ChatGPTReport a bug or request a feature. Opens a GitHub issue on the NetworkOS repo (same pipeline as the in-app feedback widget): the description is structured by an LLM, labeled (bug / enhancement / user-submitted + topical tags), and tagged with the reporter's email. Use when the user says 'file a bug', 'report a bug', 'request a feature', 'suggest an improvement', or similar.
team_search
ChatGPTSearch across your TEAMMATES' networks — if you're on a team — to discover who your colleagues know. Returns a limited view: name, company, title, LinkedIn, location, relationship warmth, and which teammate owns the relationship. Never exposes your teammates' contacts' emails, notes, or interaction history. This is a fast keyword search (matches name, company, title, headline, industry, location); use when the user asks things like "does anyone on my team know someone at Acme?" or "who at my firm knows a fintech founder?". To ask the relationship owner for a warm intro to someone you find, call request_team_intro with the contactId from a result.
trigger_sync
ChatGPTStart a contact sync: imports from Google Contacts, Gmail senders, and Calendar, then automatically dedups, enriches companies, and classifies contacts. Idempotent — if a sync is already running it won't start a second one (returns alreadyRunning). Takes a few minutes; poll get_sync_status for progress.
update_catch_up
ChatGPTSave edits to a catch-up's drafts and plan without sending and without any generation call: step subjects and bodies, reminder offsets (step2OffsetDays / step3OffsetDays, each 2 to 30 days after the previous email's actual send), the step count (1 to 3), removeStep to drop a reminder slot, and the sending account. Steps that already sent are locked server-side; while a sequence is in progress only unsent steps can change, and editing an unsent reminder's offset moves its actual send time. removeStep can't be combined with other edits. For a rewrite of the wording, use regenerate_catch_up_drafts instead.
update_company
ChatGPTEdit a portfolio company. Pass the companyId (from list_companies / search_companies) and the fields to change. Array fields (roles, currentNeeds) are replace-style — pass the full intended list. Changing website recomputes the matching domain. To remove a company, do it from the web app (soft delete).
update_contact
ChatGPTPatch profile fields on an existing contact (name, email, job title, company, LinkedIn, location, headline, phone, industry, description). Only provided fields change. Use add_contact_note for notes — not this. To mark one of the contact's addresses as wrong (it bounced, they left the company) pass emailStatus {email, status: 'invalid'}; NetworkOS then never sends to it and moves the contact to its next working address (or none). status 'active' restores it. A bounced or invalid address cannot be set as email until it is restored.
update_contact_note
ChatGPTEdit an existing contact note. Pass the noteId (from list_contact_notes) and the fields you want to change. Either content or noteType (or both) must be provided.
update_introduction_draft
ChatGPTSave edits to an intro's drafts and timing without sending, and show the updated intro card. Used by the ui://intro-confirm card's auto-save, and by you when the user dictates a change. Any of these, in one call or separately: - One email's wording: emailType + subject + body (all three together) for any draft: networkOptIn, clientOptIn, intro, the post-intro follow-ups networkFollowUp / clientFollowUp, and the no-reply reminders networkApprovalReminder / clientApprovalReminder. - Several emails at once: drafts, keyed by those same names, each {subject, body}. When you rewrite more than one email, pass them all here in ONE call so the user gets one refreshed card, not one per email. Text is saved verbatim. - Follow-up timing: networkFollowUpDays and/or clientFollowUpDays, the days after the intro email that each person's follow-up sends (whole days, 1-30, each side independent). Only before that follow-up has been sent. - No-reply reminder timing: networkApprovalReminderDays and/or clientApprovalReminderDays, the days after each opt-in email that its reminder sends if that person hasn't replied (whole days, 1-30, default 3). Only before that reminder has been sent. The result is the whole intro as saved (every draft, follow-ups, reminders), rendered as a fresh card. The auto-send jobs read the same columns, so an edit flows through to whichever step fires next. Follow-ups and reminders are fully editable here: offer this when the user asks about their timing or wording instead of sending them to the web app. To stop a follow-up entirely, use cancel_introduction_followup.
update_task
ChatGPTEdit an existing task. Pass the taskId (from list_tasks) and the fields to change. Linking is replace-style: passing contactIds/companyIds replaces the existing links (pass [] to clear). Use complete_task to mark done and archive_task to archive — not this.
upload_target_list
ChatGPTEXPERIMENT (allowlisted advisors only). A client sent the advisor a list of people they want to meet (conference attendees, prospects). Store it as a target list on that company and check who is ALREADY in the advisor's network — nothing here creates or edits a contact. Pass the list either as csvText (the raw CSV; columns are mapped automatically, any headers) or as rows (already-structured people). Returns "N of M are in your network" plus a multi-select picker of the matched people; when the user picks some and hits Next, call create_introduction_draft once per pick with this companyId — the company's saved intro template is applied automatically when it has one. Never call this for the advisor's own contacts export; it is for a CLIENT's list. If the response says the experiment isn't enabled, tell the user plainly and stop.