Production-grade editorial pipeline built with Google Apps Script, Google Sheets, and Gemini for long-form publishing workflows.
- 19-stage editorial pipeline with validation gates
- Semantic deduplication against your archive
- Automated trend and topic discovery
- Intelligent model routing
- Advisory fact-checking + risk validation
- Structured SEO/AEO/GEO metadata generation
- Editorial voice refinement workflow
- Resumable & quota-aware execution
- Production-ready Google Docs output
- Semantic memory compression
- Pipeline observability and recovery monitoring
| Feature | Included |
|---|---|
| Multi-Stage Orchestration | ✅ |
| Semantic Deduplication | ✅ |
| Trend Discovery | ✅ |
| Model Routing | ✅ |
| Fact & Risk Validation | ✅ |
| Voice Refinement | ✅ |
| SEO/AEO Metadata | ✅ |
| Google Docs Export | ✅ |
| Pipeline Recovery | ✅ |
Google Sheets orchestration dashboard showing pipeline stages, execution states, and operational controls.
Prose OS is an AI editorial operating system — a structured, software-engineered approach to long-form content creation.
It transforms single-prompt generation into a reliable, observable, and recoverable editorial pipeline.
Google Sheets acts as a lightweight state machine where each row progresses through explicit editorial states, validation layers, and recovery paths.
Instead of asking one model to do everything in one shot, Prose OS separates editorial responsibilities into independent stages:
- Insight generation
- Narrative structure planning
- Draft generation
- Fact validation
- Editorial refinement
- Formatting & metadata
- Memory & deduplication
This produces:
- Higher consistency
- Better structural quality
- Easier debugging
- Controlled execution
- More reliable long-form output
The system automatically discovers and filters high-potential topics from:
- Hacker News
- Google Trends
Ideas are filtered through configurable editorial relevance logic before entering the Idea Bank.
Once a topic is approved, it flows through a deterministic 19-stage pipeline:
- Duplicate Check — Semantic similarity scan
- Insight Generator — Core thematic insights
- Structure Planner — Editorial structure generation
- Hook Writer — Opening and framing
- Writer Part 1 — First half draft
- Writer Part 2 — Second half draft
- Word Count Gate — Quality threshold validation
- Fact Checker (Part 1) — Advisory review
- Fact Checker (Part 2) — Advisory review
- Merge & Integrity — Structural validation
- Voice Architect (Part 1) — Editorial refinement
- Voice Architect (Part 2) — Editorial refinement
- Fact Validator — Risk detection
- Link Injector — Internal linking context
- Blog Formatter — Structured blog formatting
- SEO Generator — Titles, FAQs, metadata
- Image Prompt Architect — Editorial image prompts
- Semantic Summary — Memory compression
- Final Editor — Google Doc generation & archiving
Every stage is:
- Resumable
- Observable
- Recoverable
- Validation-aware
High-level workflow showing editorial discovery, orchestration, validation, refinement, and publishing stages.
Different model tiers are used for different workloads:
- Lightweight models → Validation, formatting, metadata, deduplication
- Reasoning-heavy models → Drafting, refinement, long-form generation
Validation layers prevent malformed or incomplete output from progressing downstream.
Checks include:
- Word count thresholds
- Continuation markers
- Structural validation
- Formatting validation
- Semantic coherence
Compressed semantic summaries maintain long-term awareness across published content.
This helps prevent:
- Topic overlap
- Repeated arguments
- Redundant editorial structures
The pipeline surfaces factual risks without silently rewriting content.
This preserves:
- Authorial intent
- Editorial voice
- Structural integrity
while still improving factual reliability.
| Sheet | Role |
|---|---|
| Dashboard | Main orchestration & monitoring |
| Idea Bank | Discovery queue & approvals |
| Memory | Semantic archive |
| Published Links | Internal linking database |
| Pipeline Health | Runtime diagnostics |
The system uses explicit execution states:
- Pending
- Processing
- Quota Wait
- Error
- Ready
- Ready - Review
- Content Fail
This makes failures observable and recoverable instead of silently propagating downstream.
Runtime observability dashboard showing stage distribution, quota status, processing states, and pipeline health.
- Make a copy of the spreadsheet template
- Open Apps Script and paste the code
- Add your Gemini API key in Script Properties
- Customize
STYLE_CORE,VOICE_CORE, and discovery sources - Run "Run Pipeline" from the Prose OS menu
Prose OS supports configurable:
- Editorial voice guidance
- Model routing
- Pipeline stages
- Discovery sources
- Formatting behavior
- SEO/AEO rules
- Validation strictness
Example:
const STYLE_CORE = `
Add:
- Editorial guidance
- Formatting rules
- Structural preferences
- SEO/AEO behavior
- Publication constraints
`;const VOICE_CORE = `
Add:
- Editing priorities
- Clarity rules
- Rhythm guidance
- Refinement behavior
`;Prose OS intentionally uses Apps Script and Google Sheets because they provide:
- Native Sheets + Docs integration
- Zero infrastructure management
- Transparent orchestration state
- Low operational cost
- Easy collaboration
- Built-in persistence
This keeps the system simple, observable, and maintainable.
GPTs and Projects improve individual interactions but do not provide:
- Persistent workflow state
- Multi-stage orchestration
- Validation layers
- Recovery handling
- Structured execution
Prose OS is designed as a controlled editorial system rather than a conversational interface.
Workflow tools are optimized for moving data between services, not managing long-form editorial pipelines.
Prose OS relies heavily on:
- Native Google Sheets orchestration
- Google Docs formatting
- Stateful stage progression
- Validation-aware workflows
Rebuilding those capabilities in tools like n8n would add significant complexity without solving a meaningful problem for this use case.
Not efficiently in its current form.
Apps Script executes sequentially, so additional contributors increase queue time and reduce throughput.
The system works best for:
- Individual writers
- Small editorial teams
- Low-to-medium publishing volume
This design avoids Apps Script timeout limits and improves reliability.
Processing one stage per execution:
- Prevents cascading failures
- Simplifies retries
- Improves recoverability
- Keeps execution predictable
No.
Prose OS is designed for structured editorial workflows rather than instant conversational generation.
Stages execute independently through controlled progression and scheduled execution.
Consider migrating when you need:
- Parallel execution
- High publishing throughput
- Multiple concurrent contributors
- Advanced external integrations
- Enterprise-scale orchestration
For small editorial workflows, the current architecture remains highly efficient and low-maintenance.
Yes.
The system is optimized for free-tier usage through:
- Sequential execution
- Scoped prompts
- Validation-first workflows
- Quota-aware recovery handling
Yes.
The orchestration architecture is modular and can be adapted to alternative model providers by replacing the routing and API layers.
The pipeline compares new topics against semantic summaries and publishing history rather than relying only on keyword matching.
This helps reduce conceptual overlap across published content.
Separating drafting, validation, refinement, and formatting into independent stages improves:
- Reliability
- Consistency
- Recoverability
- Structural quality
- Workflow transparency
- Long-form writers & newsletter operators
- Editorial teams
- Independent researchers & essayists
- Workflow systems builders
- AI publishing enthusiasts
- Mass-produced SEO spam
- Fully autonomous publishing
- Bulk low-quality content generation
- Short-form social automation
Prose OS prioritizes:
- Structure
- Recoverability
- Editorial consistency
- Workflow transparency
- Operational control
over raw publishing volume.
- Architecture Guide
- Setup Guide
- Customization Guide
- Error Recovery Guide
MIT — Free to use, modify, and extend.
Contributions, architectural discussions, workflow improvements, and pipeline extensions are welcome.
⭐ If you're building structured AI publishing systems, editorial orchestration tools, or long-form workflow pipelines, consider starring the repo.
Built by Mahesh Mali — Content systems thinker, published author, and workflow architect.