Workflow Templates

Who is this for? This workflow is ideal for: HR teams and recruiters seeking to streamline resume screening. Hiring managers who want quick, summarized candidate insights. Recruitment agencies handling large volumes of applicant data. Startups and small businesses looking to automate hiring without complex systems. AI and automation professionals who want to build smart HR workflows using n8n and OpenAI. What problem is this workflow solving? / Use Case Manually reviewing resumes is time-consuming, inconsistent, and prone to human bias. This workflow automates the resume intake and evaluation process—ensuring that each applicant is screened, summarized, and scored using a consistent, data-driven method. It enhances efficiency and supports better hiring decisions. What this workflow does Accepts resume submissions via form and saves files to Google Drive. Extracts key information from resumes using AI (e.g., name, contact, education, experience). Summarizes candidate qualifications into a short, readable profile. Allows HR to rate applicants and leave comments. Logs all extracted data and evaluations into a centralized Google Sheet for tracking. Setup Resume is submitted through an n8n form. The uploaded file is automatically stored in Google Drive. n8n uses OpenAI and document parsing tools to extract candidate data. Extracted information is structured and summarized using GPT. A review form is triggered for internal HR rating and notes. All data is appended to a Google Sheet for records and filtering. How to customize this workflow to your needs** Change the form tool (e.g., Typeform, Tally, or custom HTML) based on your stack. Adapt the summary prompt to align with your specific role requirements. Add filters to auto-flag top-tier candidates based on score or skills. Integrate Slack or email to notify hiring managers when top resumes are processed. Connect to your ATS if you want to push processed resumes into your recruitment system.

Summary Engage multiple, uniquely configured AI agents (using different models via OpenRouter) in a single conversation. Trigger specific agents with @mentions or let them all respond. Easily scalable by editing simple JSON settings. Overview This workflow is for users who want to experiment with or utilize multiple AI agents with distinct personalities, instructions, and underlying models within a single chat interface, without complex setup. It solves the problem of managing and interacting with diverse AI assistants simultaneously for tasks like brainstorming, comparative analysis, or role-playing scenarios. It enables dynamic conversations with multiple AI assistants simultaneously within a single chat interface. You can: Define multiple unique AI agents. Configure each agent with its own name, system instructions, and LLM model (via OpenRouter). Interact with specific agents using @AgentName mentions. Have all agents respond (in random order) if no specific agents are mentioned. Maintain conversation history across multiple turns. It's designed for flexibility and scalability, allowing you to easily add or modify agents without complex workflow restructuring. Key Features Multi-Agent Interaction:** Chat with several distinct AI personalities at once. Individual Agent Configuration:** Customize name, system prompt, and LLM for each agent. OpenRouter Integration:** Access a wide variety of LLMs compatible with OpenRouter. Mention-Based Triggering:** Direct messages to specific agents using @AgentName. All-Agent Fallback:** Engages all defined agents randomly if no mentions are used. Scalable Setup:** Agent configuration is centralized in a single Code node (as JSON). Conversation Memory:** Remembers previous interactions within the session. How to Set Up Configure Settings (Code Nodes): Open the Define Global Settings Code node: Edit the JSON to set user details (name, location, notes) and add any system message instructions that all agents should follow. Open the Define Agent Settings Code node: Edit the JSON to define your agents. Add or remove agent objects as needed. For each agent, specify: "name": The unique name for the agent (used for @mentions). "model": The OpenRouter model identifier (e.g., "openai/gpt-4o", "anthropic/claude-3.7-sonnet"). "systemMessage": Specific instructions or persona for this agent. Add OpenRouter Credentials: Locate the AI Agent node. Click the OpenRouter Chat Model node connected below it via the Language Model input. In the 'Credential for OpenRouter API' field, select or create your OpenRouter API credentials. How to Use Start a conversation using the Chat Trigger input. To address specific agents, include @AgentName in your message. Agents will respond sequentially in the order they are mentioned. Example: "@Gemma @Claude, please continue the count: 1" will trigger Gemma first, followed by Claude. If your message contains no @mentions, all agents defined in Define Agent Settings will respond in a randomized order. Example: "What are your thoughts on the future of AI?" will trigger Chad, Claude, and Gemma (based on your default settings) in a random sequence. The workflow will collect responses from all triggered agents and return them as a single, formatted message. How It Works (Technical Details) Settings Nodes: Define Global Settings and Define Agent Settings load your configurations. Mention Extraction: The Extract mentions Code node parses the user's input (chatInput) from the When chat message received trigger. It looks for @AgentName patterns matching the names defined in Define Agent Settings. Agent Selection: If mentions are found, it creates a list of the corresponding agent configurations in the order they were mentioned. If no mentions are found, it creates a list of all defined agent configurations and shuffles them randomly. Looping: The Loop Over Items node iterates through the selected agent list. Dynamic Agent Execution: Inside the loop: An If node (First loop?) checks if it's the first agent responding. If yes (true path -> Set user message as input), it passes the original user message to the Agent. If no (false path -> Set last Assistant message as input), it passes the previous agent's formatted output (lastAssistantMessage) to the next agent, creating a sequential chain. The AI Agent node receives the input message. Its System Message and the Model in the connected OpenRouter Chat Model node are dynamically populated using expressions referencing the current agent's data from the loop ({{ $('Loop Over Items').item.json.* }}). The Simple Memory node provides conversation history to the AI Agent. The agent's response is formatted (e.g., AgentName:\n\nResponse) in the Set lastAssistantMessage node. Response Aggregation: After the loop finishes, the Combine and format responses Code node gathers all the lastAssistantMessage outputs and joins them into a single text block, separated by horizontal rules (---), ready to be sent back to the user. Benefits Scalability & Flexibility:** Instead of complex branching logic, adding, removing, or modifying agents only requires editing simple JSON in the Define Agent Settings node, making setup and maintenance significantly easier, especially for those managing multiple assistants. Model Choice:** Use the best model for each agent's specific task or persona via OpenRouter. Centralized Configuration:** Keeps agent setup tidy and manageable. Limitations Sequential Responses:** Agents respond one after another based on mention order (or randomly), not in parallel. No Direct Agent-to-Agent Interaction (within a turn):* Agents cannot directly call or reply to each other *during the processing of a single user message. Agent B sees Agent A's response only because the workflow passes it as input in the next loop iteration. Delayed Output:* The user receives the combined response only *after all triggered agents have completed their generation.

Description: This template facilitates the transfer of a folder, along with all its files and subfolders, within a Nextcloud instance. The Nextcloud user must have access to both the source and destination folders. While Nextcloud allows folder movement, complications may arise when dealing with external storage that has rate limits. This workflow ensures the individual transfer of each file to avoid exceeding rate limits, particularly useful for setups involving external storage with rate limitations. How it works: Identify all files and subfolders within the specified source folder. Recursive search within subfolders for additional files. Replicate the folder structure in the target folder. Individually move each identified file to the corresponding location in the target folder. Set up steps: Set Nextcloud credentials for all Nextcloud nodes involved in the process. -Edit the trigger settings. Detailed instructions can be found within the respective trigger configuration. Initiate the workflow to commence the folder transfer process. Help If you need assistance with applying this template, feel free to reach out to me. You can find additional information about me and my services here. => https://nicokowalczyk.de/links I have also produced a video where I explain the workflow and provide an example. You can find this video over here. https://youtu.be/K1kmG_Q_jRk Cheers. Nico Kowalczyk

This workflow creates a documentation system for n8n instances using Docsify.js. It serves a dynamic documentation website that allows users to: View an overview of all workflows in a tabular format Filter workflows by tags Access automatically generated documentation for each workflow Edit documentation with a live Markdown preview Visualize workflow structures using Mermaid.js diagrams > 📺 Check out the short 2-min demonstration on LinkedIn. Don't forget to connect! 🔧 Key Components Main Documentation Portal Serves a Docsify-powered website Provides a navigation sidebar with workflow tags Displays workflow status, creation date, and documentation links Documentation Generator Uses GPT model to auto-generate workflow descriptions Creates Mermaid.js diagrams of workflow structures Maintains consistent documentation format Live Editor Split-screen Markdown editor with preview Real-time Mermaid diagram rendering Save/Cancel functionality ⚙️ Technical Details Environment Setup Requires write access to the specified project directory Uses environment variables for n8n instance URL configuration Implements webhook endpoints for serving documentation ⚠️ Security Considerations > Note: The current implementation doesn't include authentication for editing. Consider adding authentication for production use. Dependencies Docsify.js for documentation rendering Mermaid.js for workflow visualization OpenAI GPT for documentation generation 🔍 Part of the n8n Observability Series This workflow is part of a broader series focused on n8n instance observability. Check out these related workflows: Workflow Dashboard - Get comprehensive analytics of your n8n instance Visualize Your n8n Workflows with Mermaid.js - Create beautiful workflow visualizations Each workflow in this series helps you better understand and manage your n8n automation ecosystem!

Learn how to build an MCP Server and Client in n8n with official nodes. > ⚠ Requires n8n version 1.88.0 or higher. In this example, we use Google Calendar and custom functions as two separate MCP Servers, demonstrating how to integrate both native and custom tools. How it works The AI Agent connects to two MCP Servers. Each MCP Trigger (Server) generates a URL exposing its tools. This URL is used by an MCP Client linked to the AI Agent. Whenever you make changes to the tools, there’s no need to modify the MCP Client. It automatically keeps the AI Agent informed on how to use each tool, even if you change them over time. That’s the power of MCP 🙌 Who is this template for Anyone looking to use MCP with their AI Agents. How to set up Instructions are included within the workflow itself. Check out my other templates 👉 https://n8n.io/creators/solomon/

📈 Get daily and on-demand Tesla (TSLA) trading signals via Telegram—powered by GPT-4.1 and real-time market data. This is the central AI supervisor that orchestrates seven sub-agents for technical analysis, price pattern recognition, and news sentiment. Reports are delivered in structured Telegram-ready HTML, optimized for traders seeking fast, intelligent decision-making signals. ⚠️ This master agent requires 7 connected sub-workflows to function. One of them, the News & Sentiment Agent, also requires a DeepSeek Chat API key for language processing. 🔌 Required Sub-Workflows You must download and publish the following workflows: Tesla Financial Market Data Analyst Tool Tesla News and Sentiment Analyst Tool (Requires DeepSeek Chat API Key) Tesla 15min Indicators Tool Tesla 1hour Indicators Tool Tesla 1day Indicators Tool Tesla 1hour & 1day Klines Tool Tesla Quant Technical Indicators Webhooks Tool (Requires Alpha Vantage Premium API Key) 📍 See all tools at: 🔗 https://n8n.io/creators/don-the-gem-dealer/ 🔍 What This Agent Does Listens to your trading query via Telegram Calls the Financial Analyst and News & Sentiment Analyst These agents aggregate: RSI, MACD, BBANDS, SMA, EMA, ADX Candlestick pattern + volume divergence analysis News summaries and sentiment scoring via DeepSeek Chat GPT-4.1 composes the final structured TSLA trade report with: Spot and leverage setups Signal rationale Confidence score Sentiment tag News summary 🧠 Output Example TSLA Trading Report (Daily Summary) Spot Trade • Action: Buy • Entry: 172.45 • TP: 182.00 • SL: 169.80 • Signal: RSI bounce + Bullish Engulfing • Sentiment: Neutral Leveraged Position • Position: Long • Leverage: 3x • TP: 186 • SL: 170 • Confidence: High (83/100) 📰 Top News • Tesla Model Y delivery surge – Electrek • Options market pricing in upside – Bloomberg • FSD delayed in Canada – TeslaNorth 🛠️ Setup Instructions 1. Import All 8 Workflows Ensure all sub-agents above are published in your n8n instance. 2. Create Your Telegram Bot Use @BotFather to generate the token and connect to the trigger/send nodes. 3. Connect OpenAI GPT-4.1 Add your OpenAI credentials for GPT-4.1 in the designated node. 4. Add DeepSeek Chat API Key Sign up at https://deepseek.com and insert your DeepSeek Chat credentials in the News Agent. 5. Add Alpha Vantage Premium API Key Sign up at https://www.alphavantage.co/premium/ Use HTTP Header Auth for webhook-based indicator fetchers. 6. Replace Telegram ID Update the placeholder <<replace your ID here>> with your actual Telegram numeric ID in the auth node. 📌 Included Sticky Notes ✅ Telegram Bot Setup ✅ Agent Routing & Memory ✅ Financial vs. Sentiment Trigger Flow ✅ Report Formatting (HTML) ✅ API Requirements (GPT-4.1, DeepSeek, Alpha Vantage) ✅ Troubleshooting & Licensing 🧾 Licensing & Attribution © 2025 Treasurium Capital Limited Company Architecture, prompts, and trade report structure are IP-protected. No unauthorized rebranding permitted. 🔗 For support: LinkedIn – Don Jayamaha 🚀 Deploy the Tesla Quant Trading AI system with GPT-4.1, DeepSeek Chat, and Alpha Vantage Premium—right into Telegram. All 8 workflows are required. 🎥 Tesla Quant AI Agent – Live Demo Experience the power of the Tesla Quant Trading AI Agent in action.

📊 This AI sub-agent aggregates Tesla (TSLA) trading signals across multiple timeframes using real-time technical indicators and candlestick behavior. It is a core component of the Tesla Quant Trading AI system. Powered by GPT-4.1, it consolidates 15-minute, 1-hour, and 1-day indicators, adds candlestick pattern data, and produces a unified JSON signal for downstream use by the master agent. ⚠️ This agent is not standalone. It is triggered by the Tesla Quant Trading AI Agent via Execute Workflow. 🧠 Requires: 4 connected sub-agents and Alpha Vantage Premium API Key 🔌 Required Sub-Workflows To use this workflow, you must install: Tesla 15min Indicators Tool Tesla 1hour Indicators Tool Tesla 1day Indicators Tool Tesla 1hour and 1day Klines Tool Tesla Quant Technical Indicators Webhooks Tool (provides Alpha Vantage data) 🧠 What This Agent Does Fetches pre-cleaned 20-point JSON outputs from the 4 sub-agents listed above Analyzes each timeframe individually: 15m: momentum and short-term setups 1h: confirmation of emerging trends 1d: macro positioning and trend alignment Klines: candlestick reversal patterns and volume divergence Generates a structured final signal in JSON with: Trading stance: Buy, Sell, Hold, or Cautious Confidence score (0.0–1.0) Multi-timeframe indicator breakdown Candlestick and volume divergence annotations 📋 Sample Output { "summary": "TSLA momentum is weakening short-term. 1h MACD shows bearish crossover, RSI declining. 1d candles confirm potential reversal setup.", "signal": "Cautious Sell", "confidence": 0.81, "multiTimeframeInsights": { "15m": { "RSI": 68.3, "MACD": { "macd": 0.53, "signal": 0.61 }, ... }, "1h": { "RSI": 65.0, "MACD": { "macd": -0.32, "signal": 0.11 }, ... }, "1d": { "BBANDS": { ... }, ... }, "candlestickPatterns": { "1h": "Doji", "1d": "Bearish Engulfing" }, "volumeDivergence": { "1h": "Bearish", "1d": "Neutral" } } } 🛠️ Setup Instructions Import this workflow into n8n Name it: Tesla_Financial_Market_Data_Analyst_Tool Add Required API Credentials Alpha Vantage Premium (via HTTP Query Auth) OpenAI GPT-4.1 for reasoning and synthesis Link Required Sub-Agents Connect the 4 tool workflows listed above to their respective Tool Workflow nodes Connect the webhook provider for data fetches Set Up as Sub-Agent This workflow must be triggered using Execute Workflow from the parent agent Pass in: message (optional context) sessionId (used for memory continuity) 🧾 Sticky Notes Provided 📘 Tesla Financial Market Data Analyst — Core logic overview 📈 15m / 1h / 1d Tool Notes — Indicator lists + use cases 🕯️ Klines Tool Note — Candlestick and volume divergence patterns 🧠 GPT Reasoning Note — GPT-4.1 handles final synthesis 🧩 Sub-Workflow Trigger — Proper integration with parent agent 🧠 Memory Buffer — Maintains session context across evaluations 🔒 Licensing & Support © 2025 Treasurium Capital Limited Company The logic, prompt design, and multi-agent architecture are proprietary and IP-protected. For support or collaboration inquiries: 🔗 Don Jayamaha – LinkedIn 🔗 n8n Creator Profile 🚀 Unify your Tesla trading logic across timeframes—automated, AI-powered, and built for scalers and swing traders.

Manage Calendar with Voice & Text Commands using GPT-4, Telegram & Google Calendar This n8n workflow transforms your Telegram bot into a personal AI calendar assistant, capable of understanding both voice and text commands in Romanian, and managing your Google Calendar using the GPT-4 model via LangChain. Whether you want to create, update, fetch, or delete events, you can simply speak or write your request to your Telegram bot — and the assistant takes care of the rest. 🚀 Features Voice command support using Telegram voice messages (.ogg) Transcription using OpenAI Whisper Natural language understanding with GPT-4 via LangChain Google Calendar integration: ✅ Create Events 🔁 Update Events ❌ Delete Events 📅 Fetch Events Responses sent back via Telegram 🛠️ Step-by-Step Setup Instructions 1. Create a Telegram Bot Go to @BotFather on Telegram. Send /newbot and follow the instructions. Save the Bot Token. 2. Configure Telegram Trigger Node Paste the Telegram token into the Telegram Trigger and Telegram nodes. Set updates to ["message"]. 3. Set up OpenAI Credentials Get an OpenAI API key from https://platform.openai.com Create a credential in n8n for OpenAI. This is used for both transcription and AI reasoning. 4. Set up Google Calendar In Google Cloud Console: Enable Google Calendar API Set up OAuth2 credentials Add your n8n redirect URI (usually https://yourdomain/rest/oauth2-credential/callback) Create a credential in n8n using Google Calendar OAuth2 Grant access to your calendar (e.g., "Family" calendar). ⚙️ Customization Options 🗣️ Change Language or Locale The transcription node uses "en" for English. Change to another locale if needed. ✏️ Edit Prompt You can modify the prompt in the AI Agent node to include your name, work schedule, or specific behavior expectations. 📆 Change Calendar Logic Adjust time ranges or filters in the Get Events node Add custom logic before Create Event (e.g., validation, conflict checks) 📚 Helpful Tips Make sure n8n has HTTPS enabled to receive Telegram updates. You can test the flow first using only text, then voice. Use AI memory or vector stores (like Supabase) if you want context-aware planning in the future.

🔥 Upgrade to V3 Longer blogs, Higher SEO ranking with images, charts and tables We’ve released Version 3 of our AI-Powered Blog Automation workflow. We heard your complains and made a complete redesign built for serious content creators. 📝 Read the New Articles Generated by v3 🛒 View the workflow on n8n.io ✅ Longer Blog contents 3-Agent AI Architecture as the Planner, Writer, Editor: simulate a full content team for structure, writing, and QC. A more continuous flow 📈 2x bump in SEO ranking SEO Scoring System so every article is graded on keyword density, readability, structure, backlinks, and uniqueness. IF quality doesnt meet threshold, we revise the content again. 🖼️ Smarter Visuals In-blog images via Leonardo, charts via QuickChart, tables and web scraped outbound links 🕸️ Multi-Platform Publishing Auto-posts to WordPress, Twitter (X), and Dev.to 🕵️‍♂️ Research Agent Adds quotes, stats, facts, outbound links, entities, and references to improve article credibility Content Farming V2 AI Powered Blog Automation for WordPress This workflow automatically generates and publishes 10 blog posts per day to a WordPress site. It collects tech-related news articles, filters and analyzes them for relevance, expands them with research, generates SEO-optimized long-form articles using AI, creates a matching image using Leonardo AI, and publishes them via the WordPress REST API. Every step is tracked and stored in MongoDB for reference and performance tracking. You can see the demo results for the AI based articles here: Emp0 Articles How it works A scheduler runs daily to fetch the latest news from RSS feeds including BBC, TechCrunch, Wired, MIT Tech Review, HackerNoon, and others. The RSS data is normalized and filtered to include only articles published within the past 24 hours. Each article is passed through an OpenAI-powered classifier to check for relevance to predefined user topics like AI, robotics, or tech policy. Relevant articles are then aggregated, researched, and summarized with supporting sources and citations. An AI agent generates five long-tail SEO blog title ideas, ranks them by uniqueness and performance score, and selects the top one. A blog outline is created including H1 and H2 headers, keyword targeting, content structure, and featured snippet optimization. A full-length article (1000 to 1500 words) is generated based on the outline, with analogies, citations, examples, and keyword density maintained. SEO metadata is produced including meta title, description, image alt text, slug, and a readability audit. An AI-generated image is created based on the blog theme using Leonardo AI, enhanced for emotional storytelling and visual consistency. The blog article, metadata, and image are uploaded to WordPress as a draft, the image is attached, Yoast SEO metadata is set, and the article is published. All outputs including article versions, metadata, generation steps, and final blog URLs are stored in MongoDB to allow for future analytics and feedback. Requirements To run this project, you need accounts and API access for the following: | Tool | Purpose | Notes | |--------------|------------------------------------------------------------------|-----------------------------------------------------------------------| | OpenAI | Used for blog classification, generation, summarization, SEO | Around $0.20 per day, using GPT-4o-mini. Estimated monthly: $6 | | MongoDB | Stores data flexibly including drafts, titles, metadata, logs | Free tier on MongoDB Atlas offers 512 MB, enough for 64,000 articles | | Leonardo AI | Generates featured images for blog articles | $9 for 3500 credits, $5 monthly top-up needed for 300 images | | WordPress | Final publishing platform via REST API | Hosted on Hostinger for $15/year including domain | Setup Instructions Import the provided JSON file into your n8n instance. Configure these credentials in n8n: OpenAI API key MongoDB Atlas connection string HTTP Header Auth for Leonardo AI WordPress REST API credentials Modify the classifier and prompt nodes to reflect your preferred content themes. Adjust scheduler nodes if you want to change post frequency or publishing times. Run the n8n instance continuously using Docker, PM2, or hosted automation platform. Cost Estimate | Component | Daily Usage | Monthly Cost Estimate | |---------------|------------------------------|------------------------| | OpenAI | 10 posts per day | ~$6 | | Leonardo AI | 10 images per day (15 credits each) | ~$14 (9 base + 5 top-up) | | MongoDB | Free up to 512 MB | $0 | | WordPress | Hosting and domain | ~$1.25 | | Total | | ~$21/month | Observations and Learnings This system can scale daily article publishing with zero manual effort. However, current limitations include inconsistent blog length and occasional coherence issues. To address this, I plan to build a feedback loop within the workflow: An SEO Commentator Agent will assess keyword strength, structure, and discoverability. An Editor-in-Chief Agent will review tone, clarity, and narrative structure. Both agents will loop back suggestions to the content generator, improving each draft until it meets human-level standards. The final goal is to consistently produce high-quality, readable, SEO-optimized content that is indistinguishable from human writing.

Who is this for? Marketing, content, and enablement teams that need a quick, human-readable summary of every new video published by the YouTube channels they care about—without leaving Slack. What problem does this workflow solve? Manually checking multiple channels, skimming long videos, and pasting the highlights into Slack wastes time. This template automates the whole loop: detect a fresh upload from your selected channels → pull subtitles → distill the key take-aways with GPT-4o-mini → drop a neatly-formatted digest in Slack. What this workflow does Schedule Trigger fires every 10 min, then grabs a list of YouTube RSS feeds from a Google Sheet. HTTP + XML fetch & parse each feed; only brand-new videos continue. YouTube API fetches title/description, RapidAPI grabs English subtitles. Code nodes build an AI payload; OpenAI returns a JSON summary + article. A formatter turns that JSON into Slack Block Kit, and Slack posts it. Processed links are appended back to the “Video Links” sheet to prevent dupes. Setup Make a copy of this Google Sheet and connect a Google Sheets OAuth2 credential with edit rights. Slack App: create → add chat:write, channels:read, app_mention; enable Event Subscriptions; install and store the Bot OAuth token in an n8n Slack credential. RapidAPI key for https://yt-api.p.rapidapi.com/subtitles (300 free calls/mo) → save as HTTP Header Auth. OpenAI key → save in an OpenAI credential. Add your RSS feed URLs to the “RSS Feed URLs” tab; press Execute Workflow. How to customise Adjust the schedule interval or freshness window in “If newly published”. Swap the OpenAI model or prompt for shorter/longer digests. Point the Slack node at a different channel or DM. Extend the AI payload to include thumbnails or engagement stats. Use-case ideas Product marketing**: Instantly brief sales & CS teams when a competitor uploads a feature demo. Internal learning hub**: Auto-summarise conference talks and share bullet-point notes with engineers. Social media managers**: Get ready-to-post captions and key moments for re-purposing across platforms.

This n8n template demonstrates one approach to achieve a more natural and less frustration conversations with AI agents by reducing interrupts by predicting the end of user utterances. When we text or chat casually, it's not uncommon to break our sentences over multiple messages or when it comes to voice, break our speech with the odd pause or umms and ahhs. If an agent replies to every message, it's likely to interrupt us before we finish our thoughts and it can get very annoying! Previously, I demonstrated a simple technique for buffering each incoming message by 5 seconds but that approach still suffers in some scenarios when more time is needed. This technique has no arbitrary time limit and instead uses AI to figure out when its the agent's turn based on the user's message, allowing for the user to take all the time they need. How it works Telegram messages are received but no reply is generated for them by default. Instead they are sent to the prediction subworkflow to determine if a reply should be generated. The prediction subworkflow begins by checking Redis for the current user's prediction session state. If this is a new "utterance", it kicks off the "predict end of utterance" loop - the purpose of which is to buffer messages in a smart way! New users message can continue to be accepted by the workflow until enough is collected to allow our prediction classifier to determine the end of the utterance has been reached. The loop is then broken and the buffered chat messages are combined and sent to the AI agent to generate a response and sent to the user via the telegram node. The prediction session state is then deleted to signal the workflow is ready to start again with a new message. How to use This system sits between your preferred chat platform and the AI agent so all you need to do is replace the telegram nodes as required. Where LLM-only prediction isn't working well enough, consider more traditional code-based checking of heuristics to improve the detection. Ideally you'll want a fast but accurate LLM so your user isn't waiting longer than they have to - at time of writing Gemini-2.5-flash-lite was the fastest in testing but keep a look out for smaller and more powerful LLMs in the future. Requirements Gemini for LLM Redis for session management Telegram for chat platform

Create a recommendation tool without hallucinations based on RAG with the Qdrant Vector database. This example is based on movie recommendations on the IMDB-top1000 dataset. You can provide your wishes and your "big no's" to the chatbot, for example: "A movie about wizards but not Harry Potter", and get top-3 recommendations. How it works a video with the full design process Upload IMDB-1000 dataset to Qdrant Vector Store, embedding movie descriptions with OpenAI; Set up an AI agent with a chat. This agent will call a workflow tool to get movie recommendations based on a request written in the chat; Create a workflow which calls Qdrant's Recommendation API to retrieve top-3 recommendations of movies based on your positive and negative examples. Set Up Steps You'll need to create a free tier Qdrant Cluster (Qdrant can also be used locally; it's open-sourced) and set up API credentials You'll OpenAI credentials You'll need GitHub credentials & to upload the IMDB Kaggle dataset to your GitHub.