Loading...
Loading...
A plug-and-play content engine. Researches, writes, reviews, and publishes — autonomously on your hardware. $0 API fees. 100% data privacy.
Free and open-source · Apache 2.0 · Premium prompts + guides available
$0
7,900+
6
50%
Five stages, three quality gates, zero manual steps. Every post goes through the same rigorous pipeline.
AI scans HackerNews, Dev.to, and your niche to find trending topics worth writing about. Deduplicates against your existing content.
Deep web research gathers context, statistics, and sources. Everything is verified — no fabricated citations.
Local LLM drafts long-form content with real code examples, specific tools, and actionable advice. Not generic AI slop.
Multi-model QA scores every draft on 7 dimensions. Programmatic validator catches hallucinations, fake quotes, and impossible claims.
Posts that pass quality gates publish with SEO metadata, AI-generated featured images, and podcast episodes. Static JSON pushed to any CDN.
Ollama for local inference. Your GPU, your data, your rules. No API costs for generation.
Three layers: prompt engineering, LLM cross-review, and deterministic pattern matching. Fake people, stats, and quotes get caught.
Static JSON, RSS, and JSON Feed pushed to any S3-compatible storage. Your frontend reads files — no API server needed.
Different LLMs review each other's work. Writer and critic never share a model. Adversarial quality by design.
One daemon manages N sites. Each site is a config row + storage bucket. Add a new site in minutes, not days.
Every setting, prompt, and threshold lives in PostgreSQL. Change behavior with a SQL UPDATE — no deploys, no restarts.
Google's spam policy targets AI content generated "without adding value for users." Poindexter is built around the opposite principle: generate more candidates, filter aggressively, publish only what passes.
Programmatic validator, LLM critic, topic delivery gate, internal consistency gate, vision gate, and preview screenshot check. Different models with different training data catch different blind spots.
Deterministic pattern matching catches fabricated people, statistics, quotes, and impossible claims. No LLM can override a regex match. Custom rules via the fact_overrides database.
Every article starts with web research and source verification. The writer receives real data points, real URLs, and verified facts — not just a bare topic string.
Half of generated drafts don't pass QA. That's the point. Speed comes from generating more candidates and filtering aggressively — not from lowering the bar.
The ultimate test: would you be embarrassed if someone realized the article was AI-generated? If no — if the content is genuinely useful regardless of how it was made — it's ready to publish.
Free
forever
$9/mo
or $89/year · 7-day free trial · cancel anytime
Stop paying per token. Stop trusting cloud APIs with your content. Run the whole pipeline on your own hardware and publish to anywhere.