N
Nice build. One suggestion for the dual-CRM sync: upsert HubSpot by email as the unique key and store the HubSpot contact ID in the Airtable log row, so retries from the Error Trigger don’t create duplicates. Also worth adding a quick dedupe check before the AI scoring step to save tokens on repeat leads.
N
Hey @Ab123, the tenancy and booking-race stuff is well handled above, so I’ll point at a different layer that can take the whole thing down no matter how clean the n8n side is: Meta’s side of WhatsApp. Your number has a quality rating, and it’s a shared blast radius. If several clinics sit on one number and a chunk of patients block or report messages at one of them, the rating drops and Meta tightens the messaging limit for every clinic on that number at once. That alone is a strong argument…
N
Late to this and the thread already covers the structure well (one intake workflow per channel, a canonical lead shape, one dedupe key, a status per enquiry). Three small things I haven’t seen mentioned that help with “stays reliable as enquiries grow”: 1. Let the owner mark “contacted” without learning a tool. viti_ps8 asked how the owner marks a lead as contacted - that’s what makes “unanswered” detectable at all. One simple way: put a link in the owner notification that points to a second…
N
Projeto: Tally Smart Alerts Sistema inteligente de notificações, integração por API e alertas contextuais em dispositivo externo 1. Apresentação do projeto O Tally Smart Alerts é um sistema de automação desenvolvido para conectar formulários da plataforma Tally a uma aplicação externa, responsável por interpretar interações autorizadas, organizar informações relevantes e emitir notificações inteligentes em dispositivos como smartwatches, celulares ou outros equipamentos conectados. A proposta é…
N
One from production: a small law firm, about 300 conversations a month, official WhatsApp API through YCloud. The partners don’t get an inbox. They reply from WhatsApp on their own phones, on the same number the bot uses: Meta’s coexistence mode lets the Cloud API and the Business app share one number, so the “can’t share the number” limit mentioned above is worth rechecking with your provider. Every reply they send from the phone is logged as a team message and pauses the bot for that…
N
On flowRunner’s question about making sure the focus keyword is actually searched: a simple fix is to have the model propose 10 to 20 candidate keywords and keep the one with real demand, instead of trusting its first pick. For free, Google Trends shows relative interest, and Keyword Planner gives exact volumes only once the Ads account has spend (ranges otherwise). If you want this inside the workflow without opening a DataForSEO account, I built an Apify actor that returns exact monthly…
N
Hi Robert, For your on-demand service app, I’d recommend keeping booking and provider-assignment state in your backend/database, with n8n handling the downstream automations. A reliable implementation would use unique booking/event IDs, transactional updates to avoid double assignment, separate payment and notification workflows, and safe retry handling. I’m Muhammad, Founder of MMBM Software Studio. We’ve built production systems involving transactional workflows, APIs, database operations,…
N
Hi everyone, I’m Thevanankumaran, founder of JTK LABS, a software studio in Sri Lanka that specializes in AI automation. I’m available for n8n and AI automation projects. What I can do n8n workflow development REST APIs, webhooks and JSON AI / LLM integrations (OpenAI and Anthropic) AI agents with tool calling Google Drive, Calendar and email automation PostgreSQL / Supabase and data processing Document and CV extraction with structured AI output Booking, notification and follow-up automations…
N
For the booking part, I’d keep the agent from writing to the calendar directly. Have it collect service, date and contact details, then pass them to a separate sub-workflow that checks availability and creates the booking, returning a clear confirmed/unavailable result for the agent to relay. That makes it much easier to test and stops the agent from inventing slots. Also store the conversation ID per customer, so a repeat message resumes the same session.
N
One thing that helped me keep these alerts useful: store a ‘last_touch’ timestamp and ‘next_action_due’ date on each deal, and have the scheduled workflow alert only on records where next_action_due is past and nobody has logged anything since. Send one daily digest per owner instead of one ping per deal, and add a ‘snoozed until’ field so people can quiet a deal on purpose. That keeps it from turning into noise.
N
Thanks @Anshul_Namdev, good correction. I was wrong to call them instance-wide; project-scoped variables override a global one with the same key, so checking Overview → Variables → Scope is the reliable test. @datanerdery3, for the class that may be the cleanest setup: give each student their own project and create their variables in that project, so they don’t collide. If they still can’t create them themselves, it’s likely their project role rather than the plan, and the version plus a…
N
Stephen3: workaround while investigating the dependency change. My suspicion is that the newer runtime is exposing a native dependency compatibility issue rather than a problem with your Dockerfile’s directory permissions. I’d start with the --legacy-peer- Hi Stephen thanks. We have tested a bit more. --legacy-peer-deps actually works. The old image used node version 24, while the newer one uses node 26. nodes@ /home/node/.n8n/nodes `-- @frontify/n8n-nodes-frontify@2.0.0 `-- n8n-workflow@2.16.0…
N
Hi Sarah, Urška from n8n Support here. Your workflow setup looks fine, and the 403 without the credential and the 422 with it tells me the request is reaching the course endpoint correctly, and it’s the Academy side that didn’t recognise the ID. There was a problem with that check on the Academy backend around 21–22 October, so could you run the node once more now? Please let me know the outcome!
N
Hi, Urška from n8n Support here. We have your email in our queue (ticket T-A12514) and a colleague is picking it up this morning, so you don’t need to open a second request. We will look into restore work through that ticket rather than here, since it involves your instance data. Meanwhile, let me know if I can help in any other way.
N
This is your bi-weekly security update from n8n. The following security advisories have been published since our last update: Severity: High | Deprecated Code Nodes Allow Code Execution in the n8n Main Process — GHSA-9p7v-45m2-wxgj Severity: High | Insecure GraphicsMagick Configuration — GHSA-f2cw-9483-7vcp Severity: High | Insufficient Isolation in Workflow Expression Evaluation — GHSA-gx3v-3x3v-f8r5 Severity: Medium | Browser Use Upload Tool Reads Arbitrary Files from the n8n Server —…
N
Yes, n8n fits this well if the booking record stays the source of truth and notification are just side effects webhook in → validate → upset booking by a stable booking_id split by event type into small workflows; assignment, payment update, custom alert, provide alert. Don’t put everything in one giant flow dedupe notification with a key like booking_id + notification_type before you send. An IF check alone isn’t enough when two webhooks hit at once, store the key and only send if the insert…
N
Raghavendra908: Our n8n Business license has expired, and the renewal/procurement process is currently under Legal review within our organization. Meanwhile, we need to upgrade the n8n platform to address identified security vulnerabilities. Could you please confirm whether we can proceed with the n8n upgrade while the Business license is expired? Additionally, please advise whether there are any licensing or technical restrictions that could affect the upgrade or our existing workflows,…
N
Ty_Dang_Duc: I am encountering an issue when using the AI Assistant (Preview) feature on my self-hosted n8n instance (n8n.tyauto.name.vn). When sending a prompt (e.g., “alo”), the chat interface suddenly stops and throws this error message: Error: Something went wrong before I could finish that response. Please try again. My Configuration: n8n Version: Self-hosted AI Assistant Provider: Self-hosted or OpenAI-compatible endpoint Base URL: https://integrate.api.nvidia.com/v1 Model:…
N
Sounds valid @Ayub_Shah! Just wanted to share my take.
N
Thanks for sharing these insights, @JohnYang! I’m using the easybits Extractor node in n8n, which my team and I developed. It uses Mistral, with Gemini as a fallback option, and the combination has worked really well across different document types. Gemini is also particularly good at handling handwritten documents. With the Mistral setup, I was even able to overcome my initial challenges with extracting data from large tables, so I can definitely see why you went with Mistral OCR as well. Keep…
N
Hey @Raghavendra908, while you wait for a response, here are some things that might help: Suggested resources (click for more details) @A_A4 - you’ve helped with similar issues before, can you take a look? Automatically suggested by n8n’s community bot. It’s a pilot - please share feedback here.
N
Hello Support Team, Our n8n Business license has expired, and the renewal/procurement process is currently under Legal review within our organization. Meanwhile, we need to upgrade the n8n platform to address identified security vulnerabilities. Could you please confirm whether we can proceed with the n8n upgrade while the Business license is expired? Additionally, please advise whether there are any licensing or technical restrictions that could affect the upgrade or our existing workflows,…
N
Hi @ibtihal I finally was able to take a look at this ticket, but I had some difficulties reproducing this issue. Based on the logs I would rule old encryption algorithms as the reason out, as the trace shows encryption was already set up when the router hung up. Leaving the router closing the connection either during login or when it got the command. So i have some follow up questions: Does the credential Test pass? (I’m referring to the green/red banner at the top of the credential modal,…
N
thanks for mentioning me, but have nothing to add to what @Anshul_Namdev and @minimolo suggested.
N
Adding to the flow above, for the two remaining symptoms: CSV arrives as one column That’s almost always the delimiter. Many notes and logging apps export with ; instead of ,, especially on phones set to a locale that uses a comma as the decimal separator (23,5 °C). In Extract From File → Extract From CSV → Options, add Delimiter and set it to ; (or \t for tab-separated files). If the first column name looks odd (battery), the file has a UTF-8 BOM; add Encoding: utf-8 or rename that key in a…
N
Hey @Anum_Allah_Rakha , On the two questions regarding account ownership and your Click to WhatsApp ad campaigns, here is how you handle this safely in production: For WABA and Meta App ownership, the client must always own the Meta Business Manager and the WhatsApp Business Account. You should never host a client’s production number inside your personal agency Business Manager. If an ad account gets flagged or their payment method expires, your entire agency account faces restrictions. The…
N
Hi @Ab123,tenancy, slot holds and pause flags are well covered above. You asked about edge cases, so here are five clinic-specific ones I haven’t seen mentioned (flow in the attached diagram): Shared phone numbers. Families often use one number for several patients. Before the bot shows or changes an appointment, confirm who is writing (date of birth r a booking reference). Keep reminders minimal: first name, date and time, nothing about the treatment. Relatives dates. “Tomorrow after 5” or…
N
Hello Ahmad, it’s exactly what is use today but for such a small setup i think it’s time to be a native feature
N
Hey @Aditya_Ujjwal , To answer your question directly, the current n8n Pinecone Vector Store node cannot write to document schema indexes. n8n relies on LangChain under the hood, which still calls the legacy vectors upsert endpoint. Pinecone blocks that endpoint on any index created with the new document schema. If the Pinecone console is not showing you the classic index toggle or keeps defaulting to integrated inference, you can bypass the web interface and create a true classic vector index…
N
Hi @skqnwk886, a lot of good ground is already covered here, so I’ll stick to the parts that decide whether this runs unattended without a surprise bill, which I haven’t seen raised yet. Give yourself a dry-run mode. You said no paid calls during verification, so bake that in as a flag rather than a discipline: a MOCK switch that short-circuits the Leonardo node with a canned response and a sample asset, so you can run the entire chain, job creation through render, verify, ACK and approval, as…