Coffee.Town — SMM on autopilot for a coffee shop with n8n
An n8n system that plans and writes posts, Reels and Stories every week, picks photos with the logo, asks for a yes in Telegram, publishes to Instagram and Facebook and reports back.
Coffee.Town is a Kyiv coffee shop with the Instagram account @coffee_town_kyiv and a Facebook page. There is no SMM manager, so the brief was for content to be planned, written and published on its own, with the owner only saying yes or no. We built an n8n system that plans the week, writes posts, Reels and Stories, picks photos, sends every publication to the owner for approval in Telegram, publishes to Instagram and Facebook and collects the stats. The first post went live on 6 August 2026.
Technology stack
n8n on the client’s own server with PostgreSQL, Redis and Caddy in Docker. Publishing goes directly through the Instagram Content Publishing API and the Facebook Pages API, since Meta Business Suite has no public API. Claude writes the strategy and an OpenAI model writes the copy, images are generated with OpenAI, stock photos and videos come from Pexels, and approval happens in Telegram. Model providers can be swapped in the settings without touching the workflows.
Plan and copy
Every Monday at 9:00 the strategist model plans the week: what goes out, when and in which format. It uses the brand brief and menu, stats for what has already been published, and rejected posts with the reasons, so it doesn’t repeat what the owner disliked. Then the copywriter writes the texts in Ukrainian to the brand rules: formal “you”, no Russianisms, no asterisks for bold, and only items that are really on the menu.
Approval in Telegram
Every day at 10:00 the owner gets Telegram cards for the nearest publications only: photo, text and four buttons — approve, rewrite, another photo or reject. On a rejection the bot asks for the reason, which goes into the next plan. Nothing is published without approval, and once a week the bot reminds the owner about cards left hanging.
Photos with the logo
The first automated posts got less reach than the owner’s own posts. The visuals were to blame: stock images went out instead of shots of the coffee shop, because the brand archive couldn’t be reached from the server. We fixed that and set up a queue of sources. First comes the closest matching shot from the 255-photo brand archive. If there isn’t one, an image is generated with the real Coffee.Town logo put on the cup or bag. Stock remains the fallback.
Publishing, stats and reports
Every 10 minutes the service checks approved cards, the Instagram limit and the access token, then publishes to Instagram and duplicates the post to Facebook. Every night post and account stats are stored in its own database, and Stories are captured every 3 hours because they only live for a day. On Sunday evening the owner gets a weekly report with recommendations. Meta tokens renew themselves, and a watchdog checks limits, model spend and stuck items every hour and messages Telegram when something is wrong.
Results
Since 6 August the system has published 81 pieces of content to Coffee.Town’s Instagram — 27 posts, 15 Reels and 39 Stories — and duplicated 42 of them to Facebook. Strategy and copy for three months cost about 9 dollars in API usage.
Summary
We built the whole system: n8n on the client’s server, the weekly plan and copy, approval in Telegram, photos from the brand archive or generated with the logo, publishing to Instagram and Facebook, daily stats, the weekly report, automatic token renewal and the watchdog.
