ZeroHuman-Agency

# ZeroHuman — Content Engine (Full Documentation)

> Open-source, self-hosted AI content operations platform (MIT license)

## Overview

ZeroHuman Content Engine is an autonomous, multi-tenant AI content operations platform that transforms how teams create, manage, and publish content across multiple channels. It combines intelligent web research, AI-powered content generation, multi-agent review systems, and automated publishing into a unified platform.

## Pipeline

```
Topics + Sources -> Research -> Scoring -> Draft -> Review -> Publish
                                                     ^
                                           Metrics feedback loop
```

## Architecture

- **Frontend:** Next.js 16 with React 19, Tailwind CSS, Supabase SSR auth
- **Backend:** Python FastAPI with structured agents, retrievers, scoring engine, orchestrator
- **Database:** Supabase (PostgreSQL) with Row Level Security for multi-tenant isolation
- **Social publishing:** Self-hosted Postiz satellite
- **Deep research:** local-deep-research Docker sidecar
- **Video:** HyperFrames + HeyGen integration

## Core Modules

### Research Pipeline
Pulls content from 10+ retriever sources in parallel:
- RSS feeds (configurable per brand)
- DuckDuckGo search
- Serper API (Google results)
- Tavily API (AI-optimized search)
- YouTube trends and video analysis
- X/Twitter monitoring
- Gmail integration
- Competitor spider
- Deep research jobs

Each item is deduplicated by URL and semantic similarity, then scored across configurable dimensions.

### Scoring Engine
Rates each research item across:
- Applicability: relevance to brand topics and audience
- Credibility: source trustworthiness
- Trend signal: momentum and timeliness
- Brand alignment: match with brand voice and values
- Anti-hype gate: filters sensationalized content

Scoring weights adapt via feedback loop from social performance metrics.

### Content Generation
Produces platform-native drafts with a Writer -> Editor pipeline:
- LinkedIn posts (professional, insight-driven)
- Twitter/X threads (concise, engaging)
- Instagram captions (visual-first, hashtags)
- Blog posts (SEO-optimized, comprehensive)
- Newsletter sections (long-form, valuable)

### Multi-Agent GOD Mode Review
Optional 4-agent review system:
1. **Critic:** Identifies weaknesses, structural issues, argument gaps
2. **Fact-Checker:** Verifies claims, flags unsupported statements
3. **Creative Enhancer:** Suggests engaging hooks, stronger openings, better CTAs
4. **Synthesis:** Combines all feedback into prioritized, actionable recommendations

### Humanizer
Post-processing agent that:
- Strips detectable AI writing patterns (hedging, filler, overqualification)
- Re-applies brand voice using gold examples from top-performing content
- Tracks humanization performance metrics

### Writing Lab
Experimental content iteration:
- Create sessions with a topic and two content approaches
- Run challenger/champion rounds
- Vote on preferred outputs
- Data-drive content style decisions

### Multi-Tenancy
True multi-tenant architecture:
- Each brand has isolated data (RLS policies on every table)
- Per-brand encrypted credential storage (API keys in brand_integrations)
- Per-brand sources, tone of voice, scoring weights, agent configuration
- Brand membership system for team access control

### Newsletter System
- Compose newsletters from approved drafts
- Multiple layout types: single-story, digest, announcement
- A/B subject line testing
- HTML preview before sending
- Multi-provider delivery: Resend, Brevo, Mailchimp
- Open/click rate tracking

### Social Publishing
Via self-hosted Postiz integration:
- Direct publish to LinkedIn, Twitter/X
- Schedule future publications
- Per-brand social account connections
- Analytics feedback from social platforms

### Memory System
Persistent memory layer:
- Hot memory (recent context)
- Semantic memory (embeddings-based retrieval)
- Memory archive (long-term storage)
- Episodic events
- Consolidation pipeline
- Auto-discovery from brand content

### Competitor Watch
- Spider-based competitor content monitoring
- Snapshot tracking over time
- Content gap identification

### Deep Research
- Long-form research via Docker sidecar (local-deep-research)
- Structured report generation
- Idea generation from research results

### Video
- HyperFrames motion graphics templates
- HeyGen talking-head integration
- Template customization per brand

## Compared to Alternatives

| Feature | ZeroHuman | Jasper | Copy.ai | Sight AI |
|---------|-----------|--------|---------|----------|
| Full content lifecycle | Yes | Yes (with Agents) | Partial | Partial |
| Multi-tenant native | Yes | Enterprise only | No | No |
| Multi-agent review | 4-agent GOD Mode | No | No | No |
| Content humanizer | Yes | Brand voice only | Brand voice only | No |
| Self-hosted / open-source | MIT license | No | No | No |
| Deep research built-in | Yes | No | No | No |
| Competitor monitoring | Yes | No | No | No |
| Feedback loop | Yes | No | No | No |
| Cost tracking | Yes | No | No | No |

## Getting Started

```bash
git clone https://github.com/davidesilver/ZeroHuman-Agency.git
cd ZeroHuman-Agency
cp .env.example .env.local
# Configure your Supabase URL, keys, and at least one LLM provider
npm install
npm run dev
```

## Links

- **GitHub:** https://github.com/davidesilver/ZeroHuman-Agency
- **License:** MIT
- **Tech stack:** Next.js 16, React 19, Python FastAPI, Supabase, Docker