← Back to projectsHow this works
Project
Content AI Agent
A live n8n agent that turns a written or spoken request into social copy, an image, or a video. This calls my real backend — try it yourself below.
Try it
3 of 3 tries leftResult
Your result will show up here.
An AI content agent, built in n8n
This is an AI content agent built in n8n, with tool-calling access to image generation, video generation, image editing, and multi-platform social copywriting. It was originally built for Telegram, then extended to also run as this live website demo through a second, independent trigger path.
Input (text/voice)
Agent reasoning
Tool call
Generation (image/video/copy)
Response
Why this took real engineering
- Async job architecture — submit-and-poll, since generation can take longer than a typical browser request allows.
- Per-visitor conversation memory isolation, so concurrent visitors don't bleed context into each other's sessions.
- Multi-provider cost optimization — swapped between several image/video generation providers based on real cost and quality tradeoffs.
- Integration bugs fixed from live execution traces, not just documentation — the kind of debugging that only shows up under real traffic.
Tech stack
- n8n — Workflow orchestration
- Google Gemini — Reasoning + transcription
- Cloudflare Workers AI — Image generation/editing
- Pollinations.ai — Video generation
- Google Drive + Sheets — Asset storage/logging
- Telegram Bot API — Original interface
- Next.js + Cloudflare Pages — This site