Vendo AI

Create an AI agent like step 1 of the app's builder. Choose its direction: purpose "sales" (default), "support" or "info". The agent starts in DRAFT and contacts nobody. Needs at least 1 credit on the balance (checked, not charged). Fails with a plan-limit error when all agent slots are used (see get_agent_plan_limits). Next: set_agent_company → generate_agent_script → channels.

Change an agent's name, description, direction (purpose), voice gender or preset face. Only the fields you pass change. Company name/website/description → set_agent_company; contacts → set_agent_contact_channels. Free.

Upload a custom avatar photo (JPG, PNG or GIF, ≤5 MB) from a public image URL or base64. Replaces the preset face. Free.

Set the meeting link, presentation link, email, phone, WhatsApp, Telegram and up to 10 custom links that the agent gives to visitors (also shown as screen cards). MERGE: only the channels you pass change; pass "" (or null for custom_links) to clear one. Write, free.

All voices for the website widget, Slide Presenter and Telegram voice: engine, gender, accent, expressive support, Ukrainian support, recommended flag, preview audio. Read-only. Use a voice_id from here in set_agent_voice / set_presentation_voice / set_telegram_voice_settings.

Takes no parameters.

Configure the voice of the agent's website widget exactly like the app's widget settings page and create/update the voice agent at the voice provider, so voice calls in the widget really work. Only the settings you pass change (the rest keep their saved values or the app defaults). voice_id is required the first time (list_voice_catalog). Takes up to ~1 min. Free to configure; widget voice calls cost 14 credits/min.

Send a message to the agent and get its reply using its current script — a text test, nobody is contacted. Needs a generated script. Use history for a multi-turn test.

Replace the script stages (rename, reorder, edit, add or remove stages) and optionally the tone, like the builder's script editor. Send the FULL list of stages you want; the full text is rebuilt from them. Get the current stages from get_agent (script_sections). Free.

Upload a material the agent learns from (PDF, DOC, DOCX, TXT, XLS, XLSX, CSV, PPT, PPTX, KEY; ≤10 MB) from a public URL or base64. Used the next time the script is generated (generate_agent_script). Free.

Remove an uploaded material from the agent (see get_agent → documents[].url). Free.

Read a public web page and return a knowledge text block about it. Costs 1 credit. NOT saved — append the result to company_description with set_agent_company if you want the agent to know it.

List the pages of a website (path, title, description, keywords) for this agent — used by the page-navigation feature. Costs 2 credits. Read-only for the agent (nothing saved).

Write AI descriptions + keywords for website pages (output of scan_agent_website). Costs 1 credit per 5 pages. Not saved.

Legacy support-script generator of the old text-widget creators. Prefer generate_agent_script with purpose "support". Needs company name + description. Costs 1 credit.

Generate the phrases the Slide Presenter uses to move between slides, in the presentation's language, and save them on the agent. Needs saved slides (set_presentation_slides).

Legacy flags on the agent (zoom_enabled, google_meet_enabled, website_embed_enabled). Channels are managed with the connection tools; this does not connect anything. Free.

Show what deep_scan_agent would do: cost (15 credits), duration (60-180 s) and what gets saved. Nothing runs. ALWAYS call first and get the user's approval.

Read the company website + LinkedIn and extract description, team, employee count, contacts, social links and pricing. Costs 15 credits, takes 60-180 s. With persist=true (default) the result is SAVED like in the app: company_description replaced, team/contacts/social/pricing stored (creates company + people records, fills blank contact channels). Waits up to ~200 s; if still running it returns a job_id — then call get_agent_job. Only call after preview_deep_scan and the user's approval.

Status and result of a deep scan or script generation job started by deep_scan_agent / generate_agent_script (kept 1 hour). Poll every 10-20 s while status is "running". Read-only.

Deep scans and script generations of this agent in the last hour (status only) — use when a job_id was lost. Read-only.

Save company facts (team, contacts, social links, pricing, employee count, auto re-scan) onto the agent — what the app does after a scan. Side effects: stores them in the agent, creates/updates the company and people records, fills BLANK contact channels (never overwrites manual ones). Not needed after deep_scan_agent with persist=true. Free.

AI-write "what the agent knows" about the company from its website (no full scan). Costs 1 credit. NOT saved — pass the text to set_agent_company (company_description).

Generate the agent's script (stages + full text + email/LinkedIn script) in the chosen tone for its direction (sales / support / info), using the company description and uploaded files. Saves it on the agent (overwrites). Costs 1 credit, takes 60-90 s. Needs company_name and company_description (set_agent_company). If it is still running after ~200 s you get a job_id → get_agent_job.

Older overview numbers of all agents (superseded by get_analytics_agents). Read-only.

Takes no parameters.

Full settings of one channel (website widget, button, presentation, Google Meet, phone, Telegram). Read-only.

Add a channel. WEBSITE_WIDGET needs website_url; new widgets get the app defaults (Modern design) plus your widget_config — then get_widget_embed_code. WEBSITE_BUTTON: just creates the channel (style it in get_widget_embed_code). PRESENTATION_WIDGET = publish the Slide Presenter (optional website_url). GOOGLE_MEET: prefer connect_google_meet. PHONE_CALL: prefer setup_phone_agent. Telegram → connect_telegram_bot; WhatsApp → connect_whatsapp_account. A second widget for the same URL updates the existing one. Free. Interactive cards in chat need design "modern"; in voice they need set_agent_voice screen_cards=true.

Change a channel: website widget look and behaviour (design, colours, button picture, invitation/auto-open, sounds, contact form and its wording, greetings, page navigation, voice view), name, website_url, Google Meet settings or phone texts. Only what you pass changes; widget texts and theme are merged. Free.

Show what removing a channel would do. Nothing is removed. ALWAYS call first and show the user before delete_agent_connection.

Remove a channel from the agent (the embed code stops working). No undo. Only after preview_delete_agent_connection and explicit user approval.

Conversations, messages, unique visitors, last activity of one channel. For widgets prefer get_connection_analytics (fresher). Read-only.

For a website widget/button/presentation channel: widget loaded, activated, chats started, voice calls started, plus estimated credits (text messages, voice minutes) for a date range. Read-only.

HTML to paste on a site for a channel: website widget (settings load from the server), website button (style it here: text, colours, radius, size — not stored server-side) or Slide Presenter (also the direct link). Read-only.

Create (or return the existing) private, unlisted test page with this website widget installed, to try it before adding it to a real site. Only the agent's creator can do this. Free.

Conversations of a widget/presentation channel with messages, mode (chat/voice), visitor name/email/phone and page. Paged (limit ≤500). Read-only.

Recording URLs (WAV), duration and start time of the voice calls in one widget conversation (up to 10 calls). Read-only.

The agent's presentation: PDF name, slides (text + presenter script; images with include_images), look, call-to-action, presenter voice, published channels and direct link. Read-only. (list_agent_presentations shows only the legacy creator.)

Save the slide deck — REPLACES all slides (add, remove, reorder, edit). Each slide: image (public URL or base64; JPG/PNG/WebP/GIF ≤5 MB; omit to keep the stored image of an existing slide id), extracted text and presenter script. PDF upload is not supported by the API: render the PDF pages to images first. Then write scripts with generate_slide_script and set_presentation_voice. Free.

AI-write what the presenter says on one slide, in the presentation's language. 1 credit (2 with image_url — the AI looks at the slide). NOT saved — put it into narrator_script with set_presentation_slides.

The Slide Presenter's own voice and model (the website widget voice is not touched). Same settings as the app; only what you pass changes. Needs saved slides. Takes up to ~1 min. Free to configure; presentation voice costs 17 credits/min.

Theme, colours, corner radius, controls, autoplay, voice on/off, chat on/off, Modern look (background, glass, font, panel side, presenter bubble, voice visual), slide transition and button texts. Only what you pass changes. Free.

The button viewers see: a lead form (fields) or a link. Only what you pass changes (fields replaces the whole field list). Leads land in list_narrator_leads. Free.

Direct link (narrator page) and the HTML snippet with the saved look. Publish it as a channel first (create_agent_connection type PRESENTATION_WIDGET). Read-only.

Presentations made with the OLD presentation creator (deprecated). The current Slide Presenter is in get_presentation. Read-only.

Slide Presenter analytics. view "summary" (default): views, presentations started, watch time, unique viewers, average duration, completion rate, recent sessions, per channel. view "period" with date_from/date_to: widget loads, starts, watch minutes, unique viewers, credits used. Read-only. (No per-slide drop-off.)

Leads from the presentation call-to-action form (form data, page, status, notes). Filter by status, page with limit/offset. Read-only.

Set a presentation lead's status (new / contacted / converted / spam) and notes. Free.

Show the lead that would be deleted. Nothing is deleted. Call first and show the user before delete_narrator_lead.

Permanently delete a presentation lead. Only after preview_delete_narrator_lead and explicit user approval.

Turn on the Google Meet channel the way the app does: returns the agent's own invite address ([email protected]). Invite that address to any Google Calendar event with a Meet link and the agent joins automatically. No Google login needed. Also saves the meeting settings. Free to connect; meeting minutes are billed.

Whether Google Meet is connected for this agent, its invite address, settings and totals. Read-only.

Auto-join, minutes before start, greeting, language, meeting mode (voice only / presentation — needs slides). Only what you pass changes. Free.

Show what disconnecting Google Meet would do. Nothing changes. Call first and show the user.

Remove the Google Meet channel from the agent and turn auto-join off. Already scheduled meetings are not cancelled (use cancel_meeting). Only after preview_disconnect_google_meet and user approval.

All meetings of this agent (newest first) with status, times, duration, whether a transcript exists. Paged. Read-only.

Meetings, minutes, completed, missed and credits used for a date range. Read-only.

Google Meet meetings of all your agents in a date range (default: next 7 days, max 62 days) — title, time, link, organizer, attendees, status, agent, has transcript. Read-only.

Agents that have Google Meet connected (the calendar's agent filter). Read-only.

Takes no parameters.

One meeting with transcript, summary, action items, join/leave time and duration. Read-only.

Transcript messages, summary and action items of a meeting (from the database or the recording service). Read-only.

Show which meeting the agent would skip. Nothing changes. Call first and show the user.

The agent will NOT join this scheduled meeting (the calendar event itself stays). Only scheduled, not yet started meetings. No undo. Only after preview_cancel_meeting and user approval.

Show whether the agent is in this live meeting and what removing it does. Nothing changes.

The agent leaves the live Google Meet call right now. Only works while the meeting is in progress. Only after preview_remove_agent_from_meeting and user approval.

Show or hide the VOICE button of the agent's Telegram bot (voice calls in the mini app). Needs a connected bot. Free.

Voice and behaviour of the Telegram bot's voice calls (separate from the website widget): the same settings as set_agent_voice plus the mini app look (visual). Only what you pass changes. Needs a connected bot. Takes up to ~1 min. Free to configure; calls are billed per minute.

Re-apply the bot's commands and menu button (for private and group chats) — use when the bot menu looks wrong. Free.

People/groups talking to the agent's Telegram bot: name, username, type, message count, last message time. Paged. Read-only.

All messages of one Telegram chat, incl. voice-call turns with recording URL and duration. Read-only.

Show the recipient and exact text of an operator message. Nothing is sent. ALWAYS call first and show the user the text.

Sends a REAL Telegram message from the bot to this person (human takeover). Only after preview_send_telegram_message and the user approved the exact text.

Chats, messages, voice calls, text messages, unique users and credits (text/voice) of the agent's Telegram bot for a date range. Read-only.

Phone numbers already on your account (number, country, type, monthly cost, per-minute rate). Buying numbers is done in the app. Read-only.

Takes no parameters.

Draft a call script and greeting from the agent's sales script for a call objective, in the agent's language. Free. Not saved — pass it to setup_phone_agent (or let setup_phone_agent generate it).

Create/update the agent's Voice Call channel like the app: attach one of YOUR numbers (list_phone_numbers), objective, script (generated if omitted), greeting, voice, pace, emotion, language, rules, inbound calls on/off, background sound, silence/reminder timings. Creates the phone voice agent at the provider and links the number. Takes up to ~1 min. Free to set up; calls cost 25 credits/min.

Show which number would be called from which, and the cost (25 credits/min, max 2 min). Nothing is dialled. ALWAYS call first.

DIALS A REAL PHONE: the agent calls to_number from the channel's number (max 2 min, 25 credits/min). Only call the user's own number, only after preview_phone_test_call and explicit approval.

Phone calls of this agent (incl. test and inbound) with transcript, recording URL, summary, outcome, sentiment, duration and credits, plus totals by outcome/sentiment. Read-only.

Calls attempted/completed/successful, minutes and credits of one Voice Call channel. Read-only.

WhatsApp accounts connected to your user (phone, name, status: connecting / ok / error / credentials / stopped). Account-level: auto-replies use your latest campaign's agent. Read-only.

Takes no parameters.

Returns a hosted link; the user opens it and scans the QR code with WhatsApp (Linked devices). Poll list_whatsapp_accounts until status is "ok". Free.

Takes no parameters.

Show the WhatsApp account that would be disconnected. Nothing changes. Call first.

Disconnect and delete the WhatsApp account; auto-replies stop and reconnecting needs a new QR scan. Only after preview_disconnect_whatsapp_account and user approval.

List classic Sales Process campaigns with status and live stats. Archived campaigns are hidden unless include_archived=true. Read-only.

Full campaign: status, settings (all wizard fields), sequence, channel paused flags, channelStatuses, today's sends, stats.totalContacts. Read-only.

Create a classic campaign in DRAFT — nobody is contacted until launch_campaign. Needs an audience (audienceId from save_audience/list_audiences, OR customAudienceId of a READY uploaded/CRM list) and should get an agentId (a campaign cannot be launched without an agent: AGENT_REQUIRED). branchId = CRM branch for replies (crm_list_branches; null = General). Settings (all optional): channels, dailyLimit, dailyEmailLimit, dailyLinkedinLimit, timezone, workingHours, excludeWeekends, campaignDurationDays, customInstructions (Communication Rules incl. form of address), emailSignature, notifyEmails, ccEmail, autoResponseEnabled, autoResponseDelayMinutes, followUpInterval, maxFollowUps, trackOpens, emailSenderMode (managed needs a paid plan), senderEmailIds, emailContent {mode, prompt, tone, variants, htmlSteps, followUps[{delayDays, prompt|subject+body}]}, linkedinContent {mode, invite_prompt, invite_tone, invite_exact_text, followup_delay_minutes, followup_exact_text}, selectedLinkedinAccountIds, linkedinAutoReplyDisabled. Ids are checked against the workspace. Free, no sending.

Change a campaign's name, description or any wizard setting (partial: send only what changes; settings are merged). agentId and branchId can be changed only while the campaign is a draft. Settings (all optional): channels, dailyLimit, dailyEmailLimit, dailyLinkedinLimit, timezone, workingHours, excludeWeekends, campaignDurationDays, customInstructions (Communication Rules incl. form of address), emailSignature, notifyEmails, ccEmail, autoResponseEnabled, autoResponseDelayMinutes, followUpInterval, maxFollowUps, trackOpens, emailSenderMode (managed needs a paid plan), senderEmailIds, emailContent {mode, prompt, tone, variants, htmlSteps, followUps[{delayDays, prompt|subject+body}]}, linkedinContent {mode, invite_prompt, invite_tone, invite_exact_text, followup_delay_minutes, followup_exact_text}, selectedLinkedinAccountIds, linkedinAutoReplyDisabled. On a running campaign, changed email steps/delays are not applied once anything was sent (settings.smartleadSequencePending). No sending, no credits.

Readiness report for launch/resume — NOTHING IS SENT: ready flag, blockers (AGENT_REQUIRED, AUDIENCE_REQUIRED, CLIENT_MAILBOX_NOT_CONNECTED, PAID_PLAN_REQUIRED for Vendo managed senders, INVALID_STATUS), warnings, real recipient count, channels, daily email/LinkedIn limits, sender mode, email steps, schedule and the LinkedIn accounts that will be activated. ALWAYS call this first and show the result to the user before launch_campaign.

Start (or resume a paused) campaign exactly like the app: starts email sending and queues LinkedIn on the selected (or all connected) accounts. Contacts receive REAL outreach in the working hours; credits are spent as messages go out. Only call AFTER the user reviewed preview_launch_campaign and approved it. Returns the campaign plus linkedin_activation.

Copy a campaign (settings + audience) into a new DRAFT named '<name> (Copy)'. Nothing is sent. Note: the copy re-uses company-based audiences only and has no CRM branch; check get_campaign of the copy before launching.

Pull the latest sent/opened/clicked/replied/bounced counters from the sending provider and store them. Max once per minute per campaign. No credits.

What the campaign is doing now (sending / idle / outside_hours / paused / queue_empty / done), today vs daily limits, 7–28-day email + LinkedIn series and the latest events. Read-only.

Email channel progress: audience size, processed, % complete, event counts, queue depth and provider status / sent today. Read-only.

Sales Process KPIs for the workspace: totals (campaigns, emails sent, opens, clicks, replies, contacts, rates) and per-campaign performance, optionally for a date range. Read-only. For SDR + campaigns combined use the SDR analytics tool.

Render the exact email a campaign would send — first email or follow-up #k — with the writing rules, tone, Communication Rules or fixed text. Pass campaign_id to use its saved settings (any field you pass overrides them), or agent_id + rules for an unsaved draft. No sending, no credits.

Render the LinkedIn connection note a campaign would send (AI from rules/tone, or fixed text with placeholders), with its length vs the 200-char limit. Pass campaign_id to use saved settings. No sending, no credits.

Render the first LinkedIn message sent after an invite is accepted (AI from the agent, or the fixed text with placeholders filled). Pass campaign_id to use saved settings. No sending, no credits.

Render a sample outreach email from an agent — no sending, no credits. Optional mode/prompt/tone/customInstructions/variant/followUp{index,delayDays,previous}/sample make it render exactly like a campaign would (same as preview_campaign_email).

Email funnel: sent, delivered, opened, clicked, replied (real / auto-reply / bounce), bounced; delivery block (delivered vs not delivered: bounced, rejected as spam, unsubscribed; sender inbox-placement test — real inbox vs spam folder is not reported by providers) and unique opened/clicked/replied. Read-only.

state=sent: emails already sent (filter by status, search by address); state=queued: scheduled/queued emails waiting to go out; state=prepared: pre-generated emails pipeline with status counts and preparation progress. Read-only.

Generate emails for every contact ahead of sending (legacy company-based audiences only; people/custom-audience campaigns prepare automatically after launch). Can take minutes. Nothing is sent. No credits.

Put prepared emails that failed validation/generation back in the queue (max 3 attempts each). Nothing is sent.

Dry-run: how many ready prepared emails would be scheduled into the campaign working hours. Nothing changes. Call before schedule_campaign_delivery.

Spread all ready prepared emails over the campaign's working-hour slots; they are then SENT for real automatically. Only after the user approved preview_schedule_campaign_delivery.

Dry-run: how many prepared emails would be sent immediately (outside the schedule). Nothing is sent.

Send up to `limit` prepared emails immediately, ignoring the schedule. REAL emails to real people. Only after the user approved preview_send_campaign_emails_now.

Email replies received by a campaign with body and classification: real (a person answered), auto_reply (out of office), bounce. Read-only.

Recipients with their open/click/reply counts, read live from the sending provider (throttled: once per 15 s per campaign). Read-only.

One sent email by email id (from list_campaign_emails / list_campaign_replies): content, recipient, related open/click/reply events, the reply and the conversation. Read-only.

Fetch the latest reply text for an email from the sending provider (when a reply shows as empty). No credits, nothing sent.

Dry-run of replying to a lead's email reply: shows recipient, subject and the exact text. Nothing is sent. Call before reply_to_campaign_email and show the text to the user.

Send the reply immediately in the lead's thread (bypasses daily limits). REAL email. Only after the user approved the exact text from preview_reply_to_campaign_email.

Conversations the AI is handling for a campaign (filter by status ai_handling|human_takeover|closed and sentiment positive|objection|referral|unsubscribe|neutral), plus counters by status and sentiment. Only threads of the API key's user. Read-only.

One conversation with all messages (AI, human and contact). Read-only.

Stop the AI from replying in this conversation; a human continues (reply_in_campaign_thread). Nothing is sent.

Mark the conversation closed (no more AI replies). Nothing is sent.

Dry-run: recipient, subject (defaults to Re: last subject) and the exact text. Nothing is sent. Call before reply_in_campaign_thread and show the text to the user.

Send the email to the contact now. REAL email. Only after the user approved preview_reply_in_campaign_thread. Consider takeover_campaign_thread first so the AI stops answering.

Monitor page data: AI lead discovery config + stats, audience growth by day, the last 100 activity events, event counts and credits (balance, spent today/total, daily budget). Read-only.

AI lead discovery settings of a campaign (null when never configured). Read-only.

Dry-run of enabling/changing automatic lead discovery: source, daily credit budget, leads/day. Nothing changes. Call before update_campaign_discovery whenever discovery is (or stays) enabled.

Turn on or adjust automatic lead discovery: it runs daily, SPENDS CREDITS up to daily_discovery_budget_credits and adds new leads to the campaign audience. Same args as preview_update_campaign_discovery; only after the user approved that preview. Disabling (enabled=false) needs no preview.

Leads found by AI lead discovery for a campaign, filterable by status (discovered, enriching, enriched, added_to_audience, rejected, duplicate, enrichment_failed). Read-only.

Paged activity events of a campaign (discovery, outreach, engagement, system, billing). Poll it instead of a live stream. Read-only.

Sending addresses available to this workspace (Vendo managed pool + own mailboxes) with warm-up level, daily limit and status; summary.emails_per_day_ready = capacity in managed mode. Managed senders require a paid plan to launch (summary.paid_plan). Use ids in settings.senderEmailIds. Read-only.

Takes no parameters.

The workspace's own connected mailboxes (status connecting|warming|ready|error|disconnected, warm-up reputation, last error). Capacity ≈ 50 emails/day per READY box. Used when settings.emailSenderMode=client_owned (classic campaigns and SDR). Read-only.

Takes no parameters.

Connect the user's OWN mailbox for sending. Requires SMTP/IMAP hosts+ports and an APP PASSWORD, which passes through this conversation: prefer asking the user to connect it in the app (Settings → Email Sending) unless they explicitly give you the password. WARM-UP STARTS IMMEDIATELY: the real mailbox sends and receives warm-up mail. The password is never returned. Gmail: smtp.gmail.com:465 / imap.gmail.com:993; Outlook: smtp.office365.com:587 / outlook.office365.com:993.

Retry the SMTP/IMAP connection of a mailbox in status error and refresh its status. No credits.

Dry-run: which client_owned campaigns are affected and how many usable mailboxes remain. Nothing changes.

Stop using the mailbox as a sender. If none remain, client_owned campaigns cannot send or launch. Only after the user approved preview_disconnect_client_mailbox.

LinkedIn accounts of the workspace and their status (ok = can send; ~70–80 invites/day each). detailed=true adds owner/role, campaigns and SDR searches using each account, 7-day activity, prospecting caps, warm-up stage, premium tier and proxy country. Read-only.

Per-account LinkedIn settings shared by every campaign and SDR process using the account: AI auto-replies (on/off, tone, instructions, pitch, booking link, reply delays, daily limit, working hours, agent, contact info), invites (on/off, tone, pitch, instructions, CTA), follow-ups (on/off, pitch, instructions, CTA) and signature (text, first_message|every_message). The first read creates defaults (pre-filled from campaign_id's agent).

Change AI auto-replies, invite / follow-up pitch, CTA, signature, reply delays/limits for ONE LinkedIn account (applies to all campaigns and SDR processes using it). Send only changed fields. To switch AI replies off for a single campaign use update_campaign settings.linkedinAutoReplyDisabled. No credits.

Sample invite, follow-up and auto-reply plus suggested pitch/signature for an account in a campaign (built from the campaign's agent). No sending, no credits.

Accounts working this campaign with per-account counters (pending, invites sent, connected, messaged, replied). Read-only.

Add a connected account to a running campaign; pending contacts are rebalanced across accounts and backfilled. It starts sending invites in the next cycles. No credits.

Stop using an account for this campaign; its pending contacts move to the remaining accounts. Already-sent invites stay.

Dry-run: which accounts would get the campaign audience queued, how many contacts and what is already queued. Nothing changes. Call before activate_campaign_linkedin.

Queue the campaign audience for LinkedIn on the given (or selected/all connected) accounts and add linkedin to the channels. On an active campaign REAL connection invites start within the next cycles. Only after the user approved preview_activate_campaign_linkedin.

Contacts in the campaign's LinkedIn queue with their status (pending, invite_sent, connected, message_sent, replied, failed, skipped), optionally for one account. Read-only.

LinkedIn counters (total, pending, invites sent, connected, messaged, replied, failed), invites sent today, effective daily limit and active accounts; optionally for one account. Read-only.

Stop the AI from messaging this LinkedIn contact (outreach id from list_campaign_linkedin_outreach or list_linkedin_conversations); pending auto-replies are cancelled. Nothing is sent.

LinkedIn chats of the workspace (optionally of one campaign and/or account): contact, last message, pending AI reply, human takeover flag, outreach status. Read-only.

All messages of one LinkedIn chat (chat_id from list_linkedin_conversations). Read-only.

Dry-run: recipient, exact text, cost (5 credits) and whether the balance suffices. Nothing is sent. Call before send_linkedin_message and show the text to the user.

Send a REAL LinkedIn message in the chat; costs 5 credits, the AI stops messaging this contact (human takeover) and the CRM lead moves to Manager Dialog. Only after the user approved preview_send_linkedin_message.

Dry-run of activating the phone channel with numbers the account already owns: numbers, objective, calls/day, window and estimated credits per call/day. Nothing is scheduled. Numbers cannot be bought through the API.

Schedule REAL outbound AI calls to every audience contact with a phone number; credits are charged per call minute. Same args as preview_activate_campaign_phone; only after the user approved that preview. Pause with pause_channel channel=phone_call.

Call counters: total, completed, active, pending, no answer, voicemail, failed, interested, meetings booked, not interested, average/total duration. Read-only.

Calls with contact, status, outcome, duration, sentiment, summary and recording URL. Read-only.

LinkedIn / Instagram accounts connected for publishing posts (Settings → Social), with status. Read-only.

Takes no parameters.

Returns a hosted URL a HUMAN must open in a browser to authorise a LinkedIn or Instagram publishing account. Give the link to the user; nothing is connected until they finish.

Dry-run of removing a publishing account. Nothing changes.

Remove the publishing account; it can no longer publish posts. Only after the user approved preview_disconnect_content_account.

List the workspace's custom audiences — uploaded CSV/XLSX lists, purchased audience sets, CRM selections — with status (processing|ready|enriching|error), contact counts and enrichment progress. These are different from campaign audiences (`list_audiences`). Read-only, free.

One custom audience: status, counts (with email/phone/LinkedIn), enrichment progress, column mapping and the first 20 contacts. Page through all contacts with `get_custom_audience_contacts`. Read-only, free.

Paged contacts of a custom audience (max 100 per page). Contacts added by Vendo SDR carry signal_context (why they were picked: reasoning, relevance score, ICP group). Read-only, free.

Create an empty, named custom audience (status ready) — then fill it with `add_custom_audience_contacts`. To create one from a file use `upload_custom_audience`. No credits.

Rename a custom audience. Reversible, no credits.

Append up to 5000 contacts to a custom audience. Deduplicated inside the audience by email, LinkedIn URL or phone (duplicates and contacts with no email/LinkedIn/phone are skipped). Returns added / skipped / total_contacts. No credits.

Edit one contact of a custom audience (only the fields you pass change). No credits.

Dry-run of removing one contact from a custom audience. Changes nothing. Show the result to the human before `delete_custom_audience_contact`.

Permanently remove one contact from a custom audience (not reversible). Optional reason (UI choices: Not relevant, Wrong contact, Already a customer, Bad data, Other) is logged to improve automatic selection. Only call after the user approved `preview_delete_custom_audience_contact`.

Dry-run: shows the audience name and how many contacts would be deleted. Changes nothing. ALWAYS call before `delete_custom_audience` and show the result.

Permanently delete a custom audience with all its contacts and source file. NOT reversible. Only call after the user approved `preview_delete_custom_audience`.

Step 1 of an upload: send a CSV/XLSX/XLS/TSV/TXT file (base64 or https URL) and get the detected headers, a suggested column mapping, sample rows, total rows and the list of target fields. Creates nothing, no credits. Then call `upload_custom_audience` (optionally with a corrected column_mapping).

Create a custom audience from a CSV/XLSX/XLS/TSV/TXT file given as base64 or an https URL. Parsing runs in the background: poll `get_custom_audience` until status is "ready". No credits (enrichment is a separate, paid step).

Download every contact of a custom audience as CSV text (filename, total_rows, csv). Can be large. Read-only, free.

Dry-run of enrichment: how many contacts still need enrichment, the maximum credits (1 per contact), estimated minutes, the current balance and whether it can start. Spends nothing. ALWAYS call first and show the cost to the human.

SPENDS CREDITS: starts background enrichment (email, phone, LinkedIn, WhatsApp lookups) of the not-yet-enriched contacts — 1 credit per contact actually enriched, charged as it runs. Only call after the user approved `preview_enrich_custom_audience`. Then poll `get_custom_audience_enrichment`; stop with `stop_custom_audience_enrichment`.

Status and progress of a running (or finished) enrichment: status, enrichment_total, enrichment_done, enrichment_progress (%), done flag. Poll every 10–30 s. Read-only.

Stop an in-progress enrichment after the current contact (stops further spending). Already-enriched contacts stay enriched and charged.

Turn a READY custom audience into a campaign audience and return its id ({ id, name, stats }) — pass it as audienceId to `create_campaign`. Sends nothing, no credits. The "Launch campaign" button of My Audiences.

Full card of one contact in a marketplace set: person, company, buying signal, channels. Contact details stay masked until the set is bought. Contact ids come from `get_audience_set_contacts`. Read-only, free.

Set the full left-to-right order of one pipeline's stages (every stage id exactly once — get them from crm_list_stages with the same branch). Applies to the whole workspace. Reversible.

Every pipeline of the workspace (id, name, is_default) — wider than crm_list_branches for members; use it to pick a destination for crm_add_lead_to_branch / crm_move_lead_to_branch. Read-only.

Takes no parameters.

One pipeline by id (id, name, is_default). Read-only.

Create an isolated pipeline (team / direction / client) with its own default stages. member_user_ids = who can see it (General stays open to all). Workspace admins/owner only. No credits.

Rename a pipeline and/or replace its member list. The General pipeline cannot be changed. Admins/owner only.

Dry-run: shows the pipeline and how many leads would fall back to General. Changes nothing. ALWAYS call before delete_crm_branch.

Delete a non-default pipeline; its leads are not deleted but return to General. NOT reversible. Admins/owner only. Only call after the user approved preview_delete_crm_branch.

Every pipeline a lead currently sits in, with its stage and owner per pipeline. Read-only.

Place the lead in another pipeline too (entry stage); it stays in its current ones. Reversible via crm_move_lead_to_branch.

Move the lead out of one pipeline into another (entry stage, or to_stage_id of the destination). Reversible.

Dry-run of crm_send_email_to_lead: shows recipient, subject and body exactly as they would be sent. Sends nothing. ALWAYS call first and show the draft to the human.

Sends a REAL email from the CRM card ("<your name> via Vendo AI", replies go to your own address) and logs it on the card. Members need the crm-contact-details permission. Only call after the user approved the exact draft from crm_preview_email_to_lead.

Import up to 1000 parsed rows as leads into the chosen pipelines (entry stage). Dedup by email/phone: an existing lead only gets its blank fields filled (never overwritten) and is added to the pipelines. Returns created / updated / skipped (+ reasons). No credits.

Turn a person from the Communications stream (crm_communication_history row) into a CRM lead in the New stage. Deduplicated: returns { lead_id, already_existed }. Needs at least one of email / linkedin_url / phone. No credits.

Files attached to a lead card (id, name, type, size, uploader, date). Read-only.

Attach one file to a lead card (base64 or https URL). Allowed: pdf, doc(x), xls(x), csv, txt, rtf, odt, ods, ppt(x), png, jpg, gif, webp, heic, zip. Stored privately; adds an activity entry. No credits.

Returns the file as base64 with its name and mime type (files up to 5 MB; larger ones only via REST). Read-only.

Dry-run: shows which file would be deleted. Changes nothing.

Permanently delete a file from a lead card. NOT reversible. Only call after the user approved preview_delete_crm_lead_document.

The caller's in-app notifications, newest first (type, subject, preview, isRead, createdAt), with unreadCount and paging. Types include MEETING_TRANSCRIPT, MEETING_NOTIFICATION, SUBSCRIPTION_*, PAYMENT_RECEIPT, LOW_CREDITS, CREDITS_ADDED, INVITATION, CUSTOM. CRM lead notifications are separate (crm_list_notifications). Read-only.

Number of unread in-app notifications (the bell badge). Read-only.

Takes no parameters.

Mark one in-app notification as read.

Mark every in-app notification of the caller as read (clears the badge).

Takes no parameters.

Remove one in-app notification from the caller's list.

External CRM connections of the workspace (Kommo, HubSpot): status, account, sync settings and counters. Tokens are never returned. Read-only.

Takes no parameters.

Status and settings of the Kommo or HubSpot connection ({connected:false} if none). Use it to poll after connect_crm_integration. Read-only.

Recent sync events (contact/lead created, errors) of the caller's own Kommo/HubSpot connection. Read-only.