What it does GEOFlow is a Laravel-based content factory for “Generative Engine Optimization” — essentially making your material legible to AI crawlers and search engines. It runs on PHP 8.2+ and PostgreSQL (pgvector for embeddings), and wires together knowledge bases, prompt libraries, scheduled AI generation tasks, editorial review queues, and multi-channel publishing to WordPress or static PHP agent sites.

The interesting bit The system doesn’t just generate; it packages. Each distribution channel gets a pre-configured PHP “Agent” bundle with sitemaps, llms.txt, Schema markup, and static pages — treating GEO as infrastructure, not metadata sprinkles. The RAG pipeline is unusually explicit: LLMs plan semantic chunk boundaries, but the actual slicing reconstructs from source text to avoid hallucinated structure.

Key highlights

• OpenAI-compatible and Gemini-native APIs, with model failover and call statistics
• Knowledge base uploads with rule-based or LLM-assisted semantic chunking, vectorized via pgvector
• Full editorial workflow: draft pool, audit gates, scheduled publishing, and remote distribution queues
• Target site packages generate static sites with AI-crawler-friendly markup (llms.txt, TXT maps, independent CSS)
• Docker Compose stack with Nginx/PHP-FPM, Redis, queue workers, and optional Reverb
• Admin UI in six languages; analytics dashboard tracks content ops, distribution health, and AI crawler trends

Caveats

• The README is thorough but entirely in Chinese; English docs exist but appear secondary
• “GEO” as a discipline remains fuzzy — the project acknowledges the risk of “information pollution” but the line between legitimate use and spam is operator-dependent
• Laravel Horizon and Reverb are mentioned as optional; actual production tuning guidance is thin

Verdict Worth a look if you’re running content operations at scale and want AI search visibility built into publishing infrastructure, not bolted on. Skip it if you’re after a lightweight blog or already committed to a headless stack you like.