AI Agent workflow'ları
518 sonuç — tümü kaynağa bağlı n8n referansı
Route WhatsApp support requests with Groq, Google Sheets, Gmail, Telegram and AiSensy
Quick overview This workflow receives incoming WhatsApp messages via an n8n webhook, loads an FAQ knowledge base from Google Sheets, and uses Groq (Llama 3.3) to classify intent and generate a reply. It sends the response through the AiSensy API, logs the conversation to Google Sheets, and alerts via Gmail and Telegram. How it works Receives a POST request from AiSensy on a WhatsApp webhook when a customer sends a message. Extracts the customer phone, name, message text, and timestamp, and ignores requests missing required fields. Loads FAQ question-and-answer rows from Google Sheets and sends them with the customer message to Groq (Llama 3.3) to classify intent, detect language and urgency, and draft a reply. Parses and validates the model’s JSON output and formats a safe fallback reply if parsing fails. Sends the generated reply back to the customer using the AiSensy WhatsApp API. Sends a Gmail alert when the intent is COMPLAINT and posts a Telegram alert when the intent is ESCALATE. Appends or updates a conversation log row in Google Sheets and returns a 200 OK webhook response. Setup Configure AiSensy to send message events to this workflow’s webhook URL (path: /whatsapp-suppor
Translate YouTube video metadata into multiple languages with OpenAI
Quick overview This workflow fetches a YouTube video’s title and description, uses OpenAI to translate them into multiple target languages, and updates the video’s YouTube localizations via the YouTube Data API, returning a YouTube Studio link to review the translations. How it works Runs manually when you execute the workflow. Reads the configured YouTube video ID, default language, target languages, and OpenAI model. Fetches the video’s current metadata (including existing localizations) from YouTube. Removes the default language from the target language list and prepares the language-code keys required for the translation output. Sends the title and description to OpenAI and parses the response as structured JSON with a translated title and description per language code. Updates the video on YouTube by writing the translated metadata into the video’s localizations via a single YouTube API request. Outputs a YouTube Studio translations URL so you can review the updated localizations. Setup Add YouTube OAuth2 credentials with permission to edit videos on the target channel. Add an OpenAI API key credential and choose a model name in the configuration. Set the video_id, default_lan
Answer support FAQs with Google Gemini, Supabase, Google Sheets and WhatsApp
Quick overview This workflow ingests FAQs from Google Sheets into a Supabase vector table using Google Gemini embeddings, then serves a webhook-based support chatbot that answers only from that FAQ knowledge base and escalates unanswered questions to a human via UltraMsg WhatsApp. How it works Runs manually to pull FAQ rows (Question/Answer) from a Google Sheets spreadsheet. Formats each row into a single text document (for example, Q: … A: …), generates embeddings with Google Gemini, and inserts the documents into a Supabase vector store table. Receives incoming chat messages via a POST webhook that includes message and sessionId. Uses a Google Gemini chat model with session-based memory to query the Supabase vector store for the most relevant FAQ matches and generates a reply strictly from the retrieved content. If the agent output indicates escalation to a human, sends the full message content to UltraMsg via an HTTP request to alert a WhatsApp number. Returns a JSON response to the webhook caller with either the FAQ-based answer or a confirmation that a human agent has been notified. Setup Add credentials for Google Gemini (PaLM) API, Supabase, and Google Sheets OAuth2. In Supa
Coach fitness and nutrition on Telegram with Claude, Google Sheets and Calendar
Quick overview This workflow turns a Telegram bot into a fitness and nutrition coach powered by Anthropic Claude, with voice-note transcription via OpenAI, user data stored in Google Sheets, and optional workout scheduling in Google Calendar. How it works Triggers when a new Telegram message arrives in your bot chat. If the message is a voice note, it downloads the audio from Telegram and transcribes it with OpenAI. Normalizes the incoming text or transcript into common fields (message, chat ID, and current date). Uses an Anthropic Claude agent with session memory to interpret the request and generate a coaching response. When needed, the agent reads and updates the user profile, logs workouts, logs meals with estimated macros, and reads/saves weekly plans in Google Sheets. When the user asks to schedule training, it creates a workout event in Google Calendar. Sends the coach’s reply back to the user in Telegram. Setup Add Telegram bot credentials (BotFather token) for receiving messages, downloading voice notes, and sending replies. Add an OpenAI API credential for audio transcription and an Anthropic credential for the Claude chat model. Add Google Sheets OAuth2 credentials and s
CDHandle LinkedIn inbox RAG replies with Unipile, Claude, Pinecone, and Airtable
Quick overview Automatically ingest your knowledge base, monitor your LinkedIn company page inbox, and reply to leads and customers with AI-drafted, source-grounded answers. Classifies every message, checks for spam, drafts replies using Claude and Pinecone RAG, and lets you auto-send or review each reply in Airtable. How it works A Google Drive trigger watches a folder for new or updated documents, downloading each file automatically as it's added. Each document is split into chunks, embedded, and upserted into a Pinecone vector index, building a searchable company knowledge base. A webhook receives every new LinkedIn message sent to your company page via Unipile, filtering out messages you sent yourself. The last few messages in the conversation are fetched and formatted into a chronological transcript for context. An AI agent classifies the message's intent (lead, support, partnership, recruiting, spam, or other) using shared conversation memory. A second AI agent drafts a reply, searching the Pinecone knowledge base for relevant facts before answering, and saves the draft to Airtable. Depending on a configurable auto-send toggle, the reply is sent immediately via Unipile, or he
Create AI music videos from a prompt with OpenAI, Kie.ai and BeatAPI
Quick overview This workflow collects a music video theme via an n8n form, uses OpenAI to generate music, image, and video prompts, then calls Kie to create an instrumental track and a Nano Banana image, and finally uses BeatAPI to render a publish-ready AI music video. How it works Receives a submission from an n8n Form Trigger with inputs like theme, genre, mood, duration, aspect ratio, and quality. Uses OpenAI (via an AI Agent with a structured output schema) to generate a Suno-style music prompt, an image prompt, a BeatAPI music-video prompt, and YouTube metadata. Sends the generated music prompt and settings to Kie to start audio generation, then polls Kie until the track is ready and validates the duration is within 10–180 seconds. Sends the generated image prompt to Kie (Nano Banana model) to create key art, then polls Kie until an image URL is returned. Creates a BeatAPI AI music video task using the generated audio URL, one or two image URLs (generated image plus optional cover image), and the BeatAPI prompt, then polls BeatAPI until the render succeeds. Returns the final video URL along with YouTube-ready title, description, tags, and provider task IDs in a single publish
Send daily AI-personalized cold emails with Google Places and OpenAI
Quick overview This workflow runs every morning to find local businesses via Google Places, pulls their website content to extract an email address, uses OpenAI to draft a personalised automation-focused outreach message, then sends it via SMTP while tracking contacts in an n8n Data Table. How it works Runs daily at 9:00 and sets the campaign parameters (rotating niche and city, sender details, daily send cap, and delay between emails). Searches Google Places for businesses matching the selected niche and location, then iterates through each returned place. Keeps only businesses with a website, fetches the website HTML, and extracts a likely contact email address plus a short text snippet. Filters out leads without an email and skips any email address already logged in the n8n Data Table. Limits the run to the configured daily cap and processes leads in batches. Uses OpenAI to generate a structured subject and body for a personalised cold email based on the business snippet and your campaign details. Sends the email via SMTP with your signature and unsubscribe note, logs the contact in the Data Table, and waits the configured delay before sending the next email. Setup Add credentia
Create talking avatar TikTok videos with Seedance, OpenAI and Blotato
Quick overview This workflow generates a talking-avatar video from a portrait image using AtlasCloud Seedance 2.0, creates a TikTok-ready title with OpenAI, and publishes the resulting video to TikTok via Blotato. How it works Starts when you manually execute the workflow in n8n. Sends the image URL, script prompt, and video settings to AtlasCloud Seedance 2.0 to generate an image-to-video talking avatar. Waits for a configured interval and polls AtlasCloud for the generation status until it returns completed/succeeded. Extracts the generated video URL from the AtlasCloud response. Uses OpenAI to produce a single TikTok title (under 60 characters) based on the provided script prompt. Uploads the video to TikTok through Blotato using the generated title and the selected privacy level, marking the post as AI-generated. Setup Create an AtlasCloud API key and add it as an HTTP Header Auth credential, then select it in both AtlasCloud HTTP Request steps. Add an OpenAI API credential and ensure the OpenAI chat model is selected for the title-generation step. Add a Blotato API credential, connect your TikTok account in Blotato, and select the correct TikTok account in the publishing step.
Create Seedance 2 image-to-video clips for TikTok and YouTube with OpenAI
Quick overview This workflow listens for a Telegram photo message, uploads the image to AtlasCloud to generate a Seedance 2.0 image-to-video clip, uses OpenAI to create a short SEO title, then publishes the video to YouTube and TikTok via Blotato and sends a confirmation back to Telegram. How it works Receives a Telegram message containing an image and uses the caption as the video prompt. Uploads the image to AtlasCloud and starts a ByteDance Seedance 2.0 image-to-video generation job with the configured duration, resolution, aspect ratio, and FPS. Polls AtlasCloud at a set interval until the generation status is completed or succeeded. Sends the prompt to OpenAI to generate a single short, optimized title for the video. Publishes the rendered video URL to YouTube through Blotato using the generated title. Publishes the same video to TikTok through Blotato and sends a Telegram message with the title and video link. Setup Connect a Telegram bot credential and send photos with captions to the bot to trigger the workflow. Create an AtlasCloud API header-auth credential (Authorization: Bearer ) and select it in the AtlasCloud upload, generate, and status HTTP requests. Add an OpenAI A
WSummarize Telegram-shared webpages with OpenAI GPT-4o
Quick overview This workflow listens for Telegram messages containing a URL, fetches and cleans the webpage content, summarizes it with an OpenAI chat model, and replies in the same Telegram chat with a formatted summary including key points and action items. How it works Triggers when a new Telegram message is received by your bot. Extracts the first http/https URL from the message text and captures the sender’s chat ID. If no valid URL is found, sends an error message back to the same Telegram chat. If a valid URL is found, fetches the webpage HTML and extracts the title and cleaned body text while removing common boilerplate elements. Truncates the extracted text to a configurable maximum length and sends the title, URL, and text to OpenAI to generate a structured JSON summary. Formats the JSON into Telegram HTML (title, source link, summary, key points, and insights/action items) and sends it back to the original chat. Setup Create a Telegram bot and add a Telegram API credential in n8n, then select it in both the Telegram Trigger and the send-message steps. Add an OpenAI API credential in n8n and select it for the OpenAI chat model used for summarization. Activate the workflow
Triage network read and commit requests with Slack, Anthropic, and pyATS
Quick overview This Slack ChatOps workflow uses AI agents and a pyATS MCP server to handle day-to-day data network housekeeping tasks: from on-demand checks and validations per device, to configuration change proposals that require human approval before commit. Read the docs here: https://cs.co/9004BEMGZm How it works A network engineer mentions the Slack bot in a channel or thread. The workflow acknowledges the request and captures the full message The request is sent to a Planning Agent connected to pyATS through read-only MCP tools. The agent determines whether the user wants operational data or a configuration change, gathers live device evidence if needed, and evaluates whether the request is safe. If the request is operational only, or if the proposed change is unsafe, the workflow replies in the same Slack thread with a clear technical summary, including the relevant device findings or the reason why no change will be made. If the request is a safe configuration change, the workflow creates one approval card per target device. Each card shows the proposed change, the pre-check evidence collected from the device, and the safety rationale so the engineer can review exactly wha
WSend weekly Vercel analytics digest emails with Groq and SMTP
Quick Overview This workflow runs every Monday, pulls the last 7 days of Vercel Web Analytics data, uses Groq (Llama 3.3 70B) to generate an HTML summary, and emails a weekly digest via SMTP, with an automatic raw-stats fallback if the AI step fails. How it works Runs every Monday at 9:00 AM based on your n8n instance timezone. Builds the reporting window and key settings (Vercel project/team IDs, site name, sender, and recipient) and generates millisecond timestamps for the last 7 days. Calls the Vercel Web Analytics Query API to fetch total page views and unique visitors, then fetches the top 5 pages and top 5 referrer hostnames for the same time range. Consolidates the three Vercel responses into a single JSON report with calculated pages-per-visit and defensive handling for missing or empty data. Sends the report to Groq Llama 3.3 70B to produce a structured HTML digest with headline, metrics, top pages, traffic sources, and a single actionable insight. If the Groq request errors, generates a plain HTML “raw numbers” version of the digest instead. Sends the final HTML digest email via SMTP using the configured sender and recipient addresses. Setup Create a Vercel API token and
Triage support tickets with Claude Sonnet, Infini Analytics, and Slack
Quick overview Automatically triage incoming support tickets with an AI agent powered by Claude Sonnet. High-confidence responses are sent instantly, low-confidence cases escalate to a human on Slack, and every step is logged in Infini Analytics for full observability. How it works A webhook receives the incoming support ticket and generates a unique execution_id to correlate every event of this run. Infini Analytics logs a START event for the execution. An AI Agent (Claude Sonnet) reads the ticket, optionally queries a knowledge base data table, and drafts a response with a confidence score. The agent logs structured ticket data via a data table tool (Infini Analytics records this as an EVENT). An IF node checks the confidence: high-confidence tickets get an automatic reply and log an EVENT; low-confidence tickets escalate to a human on Slack and log a WARNING. Infini Analytics logs the final END event, or an ERROR event if any step failed along the way. Setup Import this workflow into your n8n instance and open it in the editor. Create an Anthropic credential with your Claude API key and select it in the AI Agent node Create an Infini Analytics credential with your Bearer Token,
Create Xero AR/AP Markdown snapshots with optional OpenAI GPT summaries
Quick overview This workflow runs on a schedule to pull outstanding customer invoices and supplier bills from Xero, ranks the top 5 by amount due, and generates a Markdown AR/AP snapshot (optionally adding an OpenAI-based narrative summary). How it works Runs on a scheduled trigger (weekly by default). Sets report metadata and Xero invoice filters for receivables (ACCREC) and payables (ACCPAY) with AmountDue > 0. Fetches all matching authorised customer invoices and supplier bills from Xero. Normalizes dates and amounts, calculates days outstanding, sorts by amount due, and keeps the top 5 items for each list. Merges both lists and builds a Markdown snapshot with two tables and reporting notes. Outputs the final snapshot data (and, if enabled, sends the Markdown to OpenAI via an AI Agent to prepend a short AI summary). Setup Add a Xero OAuth2 credential and select the correct Xero tenant/organization ID used by the Xero nodes. Update the tenant/organization value in the snapshot settings (replace YOUR_XERO_TENANT_ID) and adjust the cadence label if needed. Configure the Schedule Trigger interval (weekly/monthly/quarterly/annual) to match your reporting rhythm. Optional: add an Open
Handle real estate SMS inquiries with Twilio, Google Sheets, Gemini, and Calendar
Quick overview Youtube Explanation: https://youtu.be/e_07nGFmVs0 This workflow replies to inbound Twilio SMS messages with property details and viewing scheduling help by looking up leads and properties in Google Sheets, using a Google Gemini agent with Redis chat memory, and checking/booking appointments in Google Calendar. How it works Receives an inbound SMS via Twilio. Looks up the sender’s mobile number in a Google Sheets “leads_info” spreadsheet and appends/updates the lead record if it does not exist, extracting a 4-digit property code from the message when available. Splits any saved semicolon-separated property IDs into individual lookups, fetches matching property rows from a Google Sheets “Property details” spreadsheet, and compiles a summary of found vs. missing property codes. Sends the user message, lead context (saved property IDs and email), current time, and the property-lookup results to a Google Gemini chat agent with Redis-backed conversation memory. Lets the agent optionally use tools to fetch additional property info from Google Sheets, update the lead’s email or interested property IDs in Google Sheets, and check or create viewing events in Google Calendar. S
Generate structured drama scripts from webhook requests using OpenAI and Google Sheets
Quick overview This workflow accepts a webhook request with a drama topic, genre, tone, and size constraints, uses Python-based genre classification plus OpenAI GPT‑4.1 to generate a full script as structured JSON, then returns the script to the caller and appends a summary log entry to Google Sheets. How it works Receives a POST webhook request containing the script topic, genre, tone, language, act count, and character count. Normalizes inputs by applying defaults, clamping acts to 1–5 and characters to 2–8, and adding request metadata like a request ID and timestamp. Classifies the genre in Python and prepares genre-specific guidance used to shape the script structure, then filters out requests marked as invalid. Sends the prepared parameters and genre guidance to OpenAI (GPT‑4.1) to generate a complete drama script as a single JSON object. Parses and validates the AI response as JSON, falls back to a safe wrapper if parsing fails, and enriches the result with counts for acts, scenes, and characters. Waits briefly and then returns the structured script to the webhook caller while also appending a summary row to a Google Sheets tab for tracking. Setup Configure the Webhook trigge
Send crowd safety SMS reroute alerts with OpenAI, Twilio, and Google Sheets
Quick overview This workflow ingests crowd density and incident reports via webhook or a 5‑minute schedule, assesses risk with Python thresholds, uses OpenAI to generate a short rerouting recommendation, then sends an SMS via Twilio and logs each alert to a Google Sheets heatmap tracker. How it works Triggers when a POST request hits the n8n webhook or every 5 minutes on a schedule. Normalizes incoming fields (location, crowd density, safety score, incidents, traveler contact, and preferences) and fills in defaults when values are missing. Calculates risk level in Python based on configurable thresholds for high density, incident count, and low safety score. Continues only for high-risk items and pauses briefly before the AI call. Uses OpenAI (via an AI agent) to generate a personalized rerouting suggestion under 150 words from the live data and traveler preferences. Sends the rerouting alert to the traveler via Twilio SMS and appends the alert details and reroute text to a Google Sheets log via the Google Sheets API. Setup Add an OpenAI API credential and select the model in the OpenAI Chat Model connection. Configure Twilio credentials (Account SID and Auth Token), replace YOUR_T
Handle theater ticket bookings with Anthropic, Google Sheets, SendGrid, and Twilio
Quick overview This workflow receives theater ticket booking requests via webhook, validates and normalizes them with Anthropic Claude, assigns seats and calculates pricing from a Google Sheets seat map, then logs the booking, sends confirmation via SendGrid email and Twilio SMS, and returns a confirmation JSON response. How it works Receives a booking request via an HTTP POST webhook and standardizes fields like show details, ticket count, and customer contact info. Uses Anthropic Claude to validate and normalize the booking request, then filters out requests with a low validation score. Retrieves the current SeatMap data from Google Sheets and assigns specific seat numbers if enough seats remain in the requested category. Calculates the final ticket total based on seat category pricing, bulk-discount rules, and tax. Uses Anthropic Claude to generate a personalized confirmation message, upsell line, and arrival instructions. Builds a final booking record (including a QR payload string), appends it to a Google Sheets tracker, sends the confirmation by SendGrid email and Twilio SMS, and responds to the webhook with the booking confirmation JSON. Setup Add an Anthropic API credential
Create AI image and video posts from Telegram with GPT-4o, AtlasCloud and Blotato
Quick overview This workflow receives a post idea via Telegram, uses OpenAI (GPT-4o mini) to generate platform-specific copy plus image/video prompts, creates media through AtlasCloud (GPT Image 2 and Seedance 2.0), requests approvals in Telegram, and publishes the approved content to seven social platforms via Blotato. How it works Triggers when a new Telegram message arrives with a post description. Uses OpenAI (GPT-4o mini) to generate copy for X, LinkedIn, Facebook, Threads, Instagram, TikTok, and YouTube Shorts, plus an English image prompt and video prompt. Sends the prompts to AtlasCloud to generate a square image (GPT Image 2) and a vertical short video (Seedance 2.0), polling each prediction until it completes. Sends the generated image and video back to Telegram for preview and asks for approval, automatically regenerating the declined media. Publishes the approved image to X, LinkedIn, Facebook, and Threads via Blotato and confirms completion in Telegram. Publishes the approved video to Instagram Reels, TikTok, and YouTube Shorts via Blotato (with synthetic/AI flags enabled where configured) and confirms completion in Telegram. Setup Create a Telegram bot with BotFather,
Generate AI songs from Telegram prompts with OpenAI and Suno
Quick overview This workflow listens for Telegram messages containing a song idea, uses OpenAI to expand the idea into a Suno-ready prompt, submits a generation request to the Suno API, polls until the track is ready, and then sends the finished audio and details back to the user in Telegram. How it works Triggers when a user sends a message to your Telegram bot. Captures the message text and chat ID, applies default generation settings, and sends an acknowledgment message back on Telegram. Uses OpenAI to rewrite the short idea into a concise, Suno-optimized music prompt based on your style preferences. Submits the enriched prompt to the Suno API to start generating a song. Waits for a configured interval, checks the Suno generation status, and repeats polling until the status is SUCCESS. Sends the generated audio URL to the user as a Telegram audio message and follows up with a completion message including the title and duration. Setup Create and select a Telegram Bot credential for the Telegram trigger and the Telegram send-audio/message steps. Add an OpenAI API credential for the OpenAI chat model used to enrich the prompt. Create an HTTP Header Auth credential for the Suno API
Create and publish AI video reels to Instagram from Telegram with OpenAI and AtlasCloud
Quick overview This workflow turns a Telegram message into an AI-generated vertical video using AtlasCloud Seedance 2.0, generates an Instagram-ready caption with OpenAI, publishes the reel via Blotato, and sends a Telegram confirmation with the published video URL. How it works Triggers when a new message is received by your Telegram bot. Uses OpenAI to expand the message into a short, visually descriptive 9:16 text-to-video prompt. Sends the prompt to AtlasCloud (Seedance 2.0) to generate a vertical video with the configured size, duration, and FPS. Waits and polls AtlasCloud for the prediction status until the video completes. Uses OpenAI to write a playful Instagram caption with 3–5 hashtags based on the video prompt. Publishes the generated video URL as an Instagram Reel through Blotato using the generated caption. Sends a Telegram message confirming the Reel is published and includes the final video URL. Setup Create a Telegram bot and add your Telegram credentials to both the trigger and the notification steps. Add an OpenAI API credential and select it for both prompt and caption generation. Create an AtlasCloud API key and configure an HTTP header auth credential (Authoriz
Score and triage job candidates with OpenAI o3-mini (OpenRouter) and ClickUp
Quick overview This workflow collects candidate CVs through an n8n Form, extracts text from a PDF, uses OpenRouter (OpenAI o3-mini) to score and summarize the candidate, and creates a ClickUp task with the evaluation for recruiter review. How it works Receives a CV submission through an n8n Form trigger (PDF upload plus message, captcha question, and consent). Extracts text content from the uploaded PDF CV. Validates that the CV text exists, the captcha answer is correct, and the candidate consent field is set to “I agree”. Sends the CV text and the candidate’s optional message to an OpenRouter chat model (openai/o3-mini) to generate a score, suggested best-fit role, and a short summary. Creates a new ClickUp task containing the AI output so the recruiting team can review the candidate. Setup Configure the n8n Form trigger fields and publish the form, then share the generated form URL with candidates. Add an OpenRouter API credential for the OpenRouter chat model used to score and summarize candidates. Add a ClickUp OAuth2 connection and select the target Workspace/Space/Folder/List in the ClickUp task creation step where candidates should be added.
Summarize Slack threads into Jira tickets with OpenRouter and Slack approval
Quick Overview This workflow turns a Slack slash-command request into an AI-drafted Jira ticket preview, then creates the Jira issue only after a user approves it in Slack using interactive buttons. How it works Receives a Slack slash command via a webhook and immediately responds with an ephemeral acknowledgement message. Reads configuration values (message count, Jira project key/type, and Slack context) and fetches recent messages from the Slack channel using the Slack conversations.history API. Cleans and formats the Slack messages into a chronological transcript, returning an error payload if no usable messages are found. Sends the transcript to an OpenRouter chat model (via an n8n AI agent) to generate a structured Jira ticket draft JSON. Parses the draft JSON and posts an ephemeral Slack preview containing the proposed title, type, priority, description, acceptance criteria, and Approve/Cancel buttons. Receives the Slack button interaction via a second webhook, parses the interaction payload, and either creates a Jira issue or cancels the request. Updates the Slack preview message to confirm ticket creation or show a cancellation notice. Setup Create a Slack app and configur
Vet community membership applications with Google Gemini, Telegram, and Sheets
Quick overview This workflow collects community membership applications via an n8n Form, vets each candidate with Google Gemini using Tavily web search and an Apify LinkedIn lookup, requests human approval in Telegram when needed, sends invite or rejection emails via Gmail, and logs outcomes to Google Sheets. How it works Receives a submission from an n8n Form that captures applicant details such as name, email, company, role, birthday, and additional notes. Normalizes the submitted fields and passes the candidate profile to a Google Gemini-powered AI Agent that can use Tavily web search, an Apify LinkedIn search actor, a calculator, and short-term memory. Produces a structured vetting result (including summary, decision status, and a community-fit score) using a structured output schema. If the AI decision is “Immediate approval”, sends the candidate summary to Telegram and waits for a response before continuing. If the AI decision is not “Immediate approval”, sends the summary to Telegram and asks the reviewer to reply with 1 (approve), 2 (reject), or 3 (do nothing). Sends an approval/invite email or a rejection email via Gmail based on the Telegram response, or stops without act