← Back to home
Archive 20 logo Multi-Agent Pipeline

An autonomous patent resurrection and public domain awareness engine. Crawls dormant patents, distills claims via RAG, aligns synthetic voice overlays, programmatically renders explainers, and publishes social campaigns.

The Mission

Millions of innovations expire or are abandoned each year, silently receding from public view into archival database records. **Archive 20** acts as a decentralized public-interest agent, automatically bringing these suppressed claims back to life. It systematically converts raw, dense patent filings into digestible, high-impact video explainers with word-level subtitles to educate the public and apply social pressure to corporate stakeholders.

System Architecture

The pipeline represents a highly modular, decoupled stack of specialized microservices coordinated by a central n8n orchestrator:

[n8n Webhook / Cron Trigger] │ ▼ 1. Scrape Filing ───► [Crawl4AI Microservice] (Playwright browser scraper) │ ▼ 2. Ingest Vectors ───► [Vector-Ingest Microservice] ───► [PostgreSQL + pgvector] │ ▼ 3. Analyze & Draft ──► [Flowise RAG Agent] ◄───────────► (Ornith 1.0 9B LLM) │ ▼ 4. Generate Voice ───► [Local TTS Engine] │ ▼ 5. Align Subtitles ──► [Whisper.cpp Container] (16kHz WAV timing) │ ▼ 6. Compile Video ───► [Editly Video Renderer] (Headless ffmpeg compiler) │ ▼ 7. Social Publish ───► [Social Posting API] ───► [Twitter / Mastodon]

Component Stack

n8n Orchestration

Acts as the nervous system. Directs file transfers, schedules execution loops, handles data serialization, and manages retry boundaries across containers.

Flowise RAG Brain

Runs a Conversational Retrieval QA Chain connecting the pgvector document index and a local LLM (Ornith 1.0 9B) via Tailscale to distill claims and generate script outlines.

Crawl4AI & Playwright

FastAPI scraper that dynamically renders JavaScript on USPTO, WIPO, or Google Patents page structures to retrieve clean metadata and diagram assets.

Whisper.cpp & Editly

Whisper aligns generated WAV voiceovers with millisecond-level word timings. Editly processes assets, layouts, and SRT tracks into final MP4 files.

Operational Guide & Best Practices

How to use Archive 20 efficiently:

1. Understanding Orchestration vs. Reasoning:

  • n8n is the master workflow controller. It handles all files (audio, image assets, video output) and sequences the external API calls. You should configure triggers and notifications in n8n.
  • Flowise acts purely as the cognitive agent. It does not touch files or make external scrapes. Instead, it is called as a webhook by n8n, accepts the query, performs RAG over the Postgres database, and returns the finished script.

2. Trigger Pathways (On-Demand vs. Scheduled):

  • On-Demand Mode: You can target specific patents by sending a POST request containing the target patent ID (e.g. US10123456B2 or WO2020257506A1) directly to the n8n Webhook.
  • Cron / RSS Mode: The pipeline can run on a periodic schedule (e.g., once daily) to crawl RSS feeds or patent databases for recently abandoned patents, filter them by keywords, and trigger automatically.

3. Tailscale Inference Binding:

  • The Flowise LLM nodes route through Tailscale to your local Windows workstation's Ollama instance. Ensure Tailscale is active on both the server and your workstation during pipeline runs.

Resurrected Patents Dashboard

Live registry of scraped, analyzed, and synthesized patents in the Archive 20 RAG knowledge pool. Trigger pipeline steps manually below.

Scrape & Ingest New Patent

Patent ID Title Company / Assignee Year Pipeline Status Date Resurrected Control Action
Connecting to n8n pipeline...