Last week my content pipeline published to nine platforms while I ran two coaching sessions, walked three dogs, and went to bed at a normal time.
This is not a flex. It is a description of what I was trying to build when I started ContentForge, and why it is worth writing down now that it is working.
The problem it solves: I publish across LinkedIn, Mastodon, Pinterest, Substack, Medium, Dev.to, IndieHackers, Quora, and a personal blog. I have 17 books that need regular promotion. I run six web apps. I hold a senior enablement role full-time. Manually writing platform-specific content was consuming three to four hours every week, inconsistently.
ContentForge takes a single story and produces platform-appropriate versions for each channel. A LinkedIn post reads like a LinkedIn post. A Dev.to article has technical framing a Dev.to audience wants. A Pinterest pin has the SEO structure Pinterest requires.
The technical reality: Vercel-hosted app with a Supabase backend. The AI generation takes a seed input, extracts the relevant elements (the feeling, the insight, the tension, the proof), and generates platform variants. Output goes into a marketing queue table in Supabase. A local Python process called Clawbot watches the queue and drives publication.
AI generation happens at content creation time. Publication is async and does not require inference on the publish side. This keeps the publish process simple and the AI costs bounded.
What I got wrong first: the initial version tried to generate from topics. The output was technically correct and completely lifeless. Topics are not seeds. Moments are seeds.
The second version required a specific story as input. That changes everything about output quality.
The third version, which runs now, extracts story elements before generating. The moment, the feeling, the insight, the proof, the tension. If you cannot name those elements, you do not have a seed yet.
Numbers from this week: nine platform assets generated from one seed. Total active time on content: approximately 45 minutes reviewing before publish. Previous approach: three hours for the same volume, with worse platform-native quality.
ContentForge is at contentforgehq.com. The queue architecture is straightforward. Happy to share more detail if useful.
Top comments (0)