EzDistro
Self-hosted SEO content automation and publishing pipeline
In their words
EzDistro is a self-hosted software platform that turns a list of topics into reviewed, publish-ready articles. It indexes existing WordPress posts into a vector knowledge base, retrieves relevant context, generates structured articles section by section using a chosen language model, and publishes back to WordPress only after explicit human approval. The application runs on your own infrastructure with your own AI providers.
Core Subsystems
-
Indexer: Fetches published WordPress posts, strips HTML, chunks text, embeds passages, and stores them in Qdrant with content-hash freshness checks.
-
Writer: Retrieves context, builds a schema-validated outline, generates sections independently, and computes a deterministic SEO score.
-
SEO Research Engine: Combines Google Ads keyword demand, competitor crawling, and keyword clustering to score article opportunities.
-
Image Pipeline: Plans and generates cover and interior images using configurable providers, quality gates, and WebP, AVIF, or JPEG optimization.
-
Review and Publishing: Enforces a mandatory human approval gate and executes idempotent WordPress publishing to prevent duplicate posts.
Fact file
Developer tool- Web
- Self-hosted
- WordPress
- Qdrant
- PocketBase
- Background jobs
- 14 job types
- Database collections
- 33 PocketBase collections
- External dependencies
- 0 Redis, Celery or n8n processes
What is EzDistro?
Indexing WordPress and building a vector knowledge base
Published posts are fetched from WordPress, stripped of HTML, chunked into passages, embedded, and stored in Qdrant under a project namespace. Content-hash checks skip unchanged posts during a run, and interrupted jobs resume from saved checkpoints without starting over from scratch. Every secret is encrypted at rest inside the connections tab, allowing EzDistro to maintain a secure connection to your infrastructure without relying on hosted external schedulers or third-party workflow queues.
Generating retrieval-grounded articles and managing research
Topics queued by import or individually trigger context retrieval from the vector base. An LLM generates a schema-validated outline, followed by independent section generation that builds the full draft and computes a deterministic SEO score. The search engine combines Google Ads keyword demand, competitor crawling, and keyword clustering to score article opportunities, refusing to estimate metrics it cannot observe directly. Each step relies on the exact context sent to the model, which remains fully inspectable inside the retrieval tab of the workspace.
Reviewing drafts and executing idempotent publishing
Articles arrive in a three-pane workspace showing validation reports, internal links, and revision histories before any content goes live. A mandatory human approval gate prevents automated publishing, while idempotent publishing updates the correct WordPress post without creating duplicates, uploads the planned images, and records every attempt in an append-only audit log. The entire background work runs in a standalone asyncio worker that claims jobs from PocketBase using atomic leases, ensuring that a crash or a server restart never corrupts an article or pushes unreviewed drafts to a live site.