“Keep Logic Apps / n8n as the orchestrator, SQL as the audit log of record, and put a hard gate before anything auto-closes or pages - agent proposes, human (or a second check) confirms. Biggest footgun is letting the agent write severity without a replayable evidence row.”
“We designate CVEs to an agent being tricked into using their tools in unintended ways? Hasn't that been like.. prompt injection 101 since they became a thing? The most elementary agent attack?”
“From what I understand, if I host/manage client workflows and credentials in my own n8n instance, n8n requires an Enterprise/commercial license. So the alternative seems to be having each client own their own n8n account/instance.”
“One gap not covered yet: what happens after a send. With local n8n + SMTP, you get zero visibility into bounces unless you build it yourself — a hard bounce (invalid address) will just silently fail or get lost in SMTP logs you're not watching.”
“Localhost n8n + localhost ollama
I guess these are only free tools. Well if you know, how to prompt AI then you can use AI to make your workflow also”
“You should be able to manage multiple clients on one server , rather then just building n8n workflows and using its own queue management , its better to add redis queue management so that even if things spike on some days you get the results. Since voice is handled by external api it should be easy manage couple of clients on 1 server.”
“A separate external probe should hit the n8n endpoint or platform and alert if it stops responding. That catches host freezes and outages the workflow cannot report.”
“Bland can handle the voice side, but all those calls can still turn into n8n executions, calendar writes, CRM updates and follow ups on your end. I’d size the infrastructure around those bursts instead of the average day.”
“Bland handles the voice layer for the setups I work around but there’s still all the n8n logic, failed workflows, client changes and random edge cases that become your problem. I’d make sure the retainer pays for that ongoing responsibility before worrying about how attractive the headline price looks.”
“Has anyone run into this with app-only Graph auth in n8n? Wondering if it's an n8n-specific thing (like a hidden default request property), or purely an Azure/Graph permissions thing outside of n8n's control.”
“- n8n launched Assistant to build, run, and debug workflows. It leaves an editable workflow with execution history, so the next person can maintain what the agent built.”
“I suggested that to be done via n8n or LogicApps (simply put) to pull data from SIEM, search similar cases in CMDB (data enrichment) and agent to process and produce the result with confidence score, then submit it to CMDB.”
“One blind spot worth closing, because the success rate is built on execution status: n8n marks an execution as successful in a few cases where the data never arrived.”
“I found hosting n8n on a VPS can be pretty costly. You'll have to get a more powerful instance to be able to run this smoothly. I would have the client set up their own account with your hosting service and ask them to pay it directly.”
“Yes - local n8n is enough for the core flow, no paid automation tools needed. Your list in a Google Sheet or CSV -> n8n reads it -> Gmail or SMTP node sends -> mark each row as sent.”
“Yeah locally hosted n8n handles this fine, you don't need anything paid for basic email sending off a list. Just watch your email sending limits depending on which smtp or provider you connect, gmail and most free tiers cap you pretty low daily so that's usually the real bottleneck, not n8n itself”
“I edited a video showing my hermes + n8n stack which is ultimately simple but took a bit of trial and error of different things until it came together. Let me know what you think.”
“more setup than wiring up otel, but it's the only way to catch the "green checkmark, wrong outcome" case, since n8n's own execution status has no idea what "correct" means for your specific workflow.”
“Two more that look like n8n problems and are not. Set the two-step PIN on the number after registration, otherwise sends fail with 133010 account not registered even though the dashboard shows it connected.”
“once it's actually attached, the n8n side is straightforward, just swap out the phone number id and auth token in your webhook node. the cryptic part is always getting meta to accept the number in the first place”
“You are adding human in the loop where the industry is moving in the opposite direction.This is generally true across all roles, but first impact will be working that is more trivial for an agent or n8n flow.So consider this and pick roles wisely, where you can see this existing for a bit.”
“Mine writes rows into a Google Sheet, but in your case one extra n8n node that appends a daily rollup (counts by department and seniority) to a Sheet does the same job, and Postgres stays the source. The Sheet has one tab published to the web as CSV.”
“One conversation should end with a clear next step, not a tour of n8n. Keep the offer narrow enough that a clinic can say yes without redesigning its whole operation.”
“I don’t have any issues with this part and it has been working quite fine. Right now I just get the results as a Slack digest, but it's not great for spotting trends over time.”
“Also curious about source quality control, since SearXNG is aggregating from multiple engines, does the n8n workflow do any filtering or ranking before it writes sources into the note, or does it just pull in whatever comes back first?”
“Spent a day building an email triage workflow for my own inbox and hit four n8n behaviours that are not obvious. Sharing the cut-down version and the fixes.”