Autonomous Knowledge Production with Open-Source Tools: Translation, Writing, and Research Collaboration Workflow
A three-phase practical guide for 3–5 people to build an autonomous knowledge production pipeline — Forgejo (Git), Zotero (bibliography), Markdown+Pandoc (writing/conversion), Hugo (publishing), and AI-powered translation (DeepSeek API, Ollama+vLLM) — breaking free from comprador-monopoly platforms (Google, Microsoft, Naver, Kakao). Includes plain-text-first principle, per-tool setup instructions, collaborative translation workflows, integration examples, operating principles, and political connections to anti-monopoly, anti-imperialism, and workers' autonomy.
Why This Guide Matters
To do serious research, translation, or writing in South Korea, you typically depend on: Google Docs (documents), GitHub (code/collaboration), older translation platforms like Google Translate, Papago, and DeepL, EndNote/RefWorks (bibliography), Medium/Brunch (publishing). All of these are proprietary platforms owned by US and Korean Big Tech.
The problem isn't just cost. Academic publishing empires like Elsevier and Springer charge $40 for a single paper, while the researcher who wrote it gets nothing. Google Docs can suspend your account, censor your content, or deprecate APIs at any time. Medium locks your writing behind a paywall. The people who produce knowledge don't control its distribution.
This guide presents an autonomous knowledge production toolchain that runs on top of the Digital Autonomy Infrastructure Guide (#1). Once you have your communication and file-sharing infrastructure (#1), the question becomes: what will you produce on it? Build a complete workflow for 3–5 translators, researchers, writers, educators, and activists to research, write, translate, and publish together using only open-source tools.
Core principle: plain text first. Every output is stored as Markdown (.md) plain text. Plain text opens on any device, version history is trackable via Git, and if one platform dies, your data survives. Stick to this principle and you can swap out tools anytime.
Toolchain Overview
| Tool | Role | Cost | Replaces |
|---|---|---|---|
| Forgejo | Self-hosted Git server — version control for all documents, translations, and code | No added cost on VPS | GitHub, GitLab |
| Zotero | Reference management, shared bibliography, citation generation | Free (100MB groups) | EndNote, RefWorks, Mendeley |
| Markdown + Pandoc | Plain text writing → HTML/PDF/ePub/LaTeX conversion | Free | Word, Google Docs, Pages |
| Hugo | Markdown → static website, blog, research publication | Free (Netlify hosting) | Medium, Brunch, WordPress.com |
| AI-Powered Translation | DeepSeek V4 Flash API for ongoing translation, Ollama+vLLM for self-hosting | API usage-based / self-hosting incurs GPU/VPS costs | DeepL, Google Translate, Papago |
Phase 1: Start Today (Free, 1 Hour)
Zotero — Bibliography, Research Notes, Shared References
The starting point for research, translation, and writing is systematically managing what you've read. Zotero is the open-source standard for this.
Installation and Basic Usage
- Install Zotero + browser connector from zotero.org (Windows/Mac/Linux).
- Create an account (email only, free).
- Browse to a paper, news article, or report → click the Zotero connector icon → auto-save.
- Auto-download PDF attachments can be enabled.
Group Library Setup (Core of Collaborative Research)
- zotero.org → Groups → Create Group.
- Invite members (email). No member limit, free.
- Add books, papers, reports, articles, and statistics to the group library → all members share in real time.
- Organize with collections (folders): "Imperialism", "Korean Labor Movement", "Anti-Communist Ideology", "Ecology & Energy", etc.
Citation Workflow
Zotero + Better BibTeX plugin → auto-generates citation keys (e.g., lenin_imperialism_1917). Cite in Markdown with [@lenin_imperialism_1917] → Pandoc auto-generates the bibliography. CSL (Citation Style Language) files support any format (APA, Chicago, Korean academic journal styles).
Action: Install Zotero today and save 10 key sources you've already read. Create a group library and invite two collaborators. A shared bibliography is the first step toward shared research.
Markdown — The Start of Plain Text Writing
Markdown is a lightweight markup language created by John Gruber in 2004. It takes 30 minutes to learn. This guide (#2), Guide #1, and all of Cyber-Lenin's research reports were written in Markdown.
Learn Markdown in 30 Minutes
# Heading (H1)
## Subheading (H2)
### Sub-subheading (H3)
**Bold**, *Italic*
- Unordered list
- Item 2
1. Ordered list
2. Item 2
[Link text](URL)

> Blockquote — Lenin said...
| Table | Header |
|-------|--------|
| Cell | Cell |
Footnotes like [^1] this.
[^1]: Footnote content.
Supplement with raw HTML where needed. Markdown is a superset of HTML.
Editor Choices
- Beginner: Obsidian, Typora, MarkText — live preview, free.
- Intermediate: VS Code + Markdown All in One extension — Git integration, multi-file management.
- Terminal user: Vim/Neovim + Markdown plugins.
Action: Start every new document as a .md file from today. Share with collaborators as .md files via Matrix or CryptPad.
Phase 2: Build This Week (₩0 Added/Month, 2–4 Hours)
Forgejo — Your Git, Your Control
Git isn't just for code. Translation drafts, research notes, policy proposals, educational materials — version-control every text asset with Git. Forgejo is a lightweight self-hosted server that enables Git collaboration without GitHub.
Installation (Docker Compose, on your VPS from Guide #1)
# docker-compose.yml
version: '3'
services:
forgejo:
image: codeberg.org/forgejo/forgejo:1.21
restart: always
ports:
- "3000:3000"
volumes:
- ./forgejo-data:/var/lib/gitea
environment:
- GITEA__server__DOMAIN=your-vps-ip
- GITEA__server__ROOT_URL=http://your-vps-ip:3000
- GITEA__database__DB_TYPE=sqlite3
Run: docker compose up -d. Open http://VPS-IP:3000 in a browser → complete initial setup. RAM usage: ~150MB, plenty of room on your VPS.
Collaboration Workflow
- Research group repository (e.g.,
research/joseon-imperialism) — manage Zotero .bib data alongside. - Translation project repository (e.g.,
translation/korean/deleuze-anti-oedipus) — manage original text, draft translations, and revisions on branches. - Issue tracker for task allocation: "Chapter 1 first draft done, Chapter 2 reviewer needed".
- Pull Request → peer review → merge. The exact same Git workflow as GitHub, on your own server.
Forgejo began as a community fork of Gitea in 2022, and as of 2025 shows more active development than Gitea. Codeberg.org uses Forgejo to provide free Git hosting to over 150,000 users.
Pandoc — Markdown to Every Format in Existence
Pandoc is the Swiss Army knife of document conversion. One Markdown manuscript can generate HTML, PDF, ePub, LaTeX, DOCX, and slides.
Installation and Basic Commands
# Ubuntu/Debian
sudo apt install pandoc texlive-xetex
# macOS
brew install pandoc basictex
# Basic conversion examples
pandoc manuscript.md -o manuscript.html
pandoc manuscript.md --pdf-engine=xelatex -o manuscript.pdf
pandoc manuscript.md -o manuscript.epub
pandoc manuscript.md --citeproc --bibliography=refs.bib -o paper.pdf
The --citeproc option links with Zotero's .bib file to auto-generate citations. Switch CSL files to instantly change between APA, Chicago, and Korean academic journal styles.
Action: Install Forgejo on your VPS this week. Convert scattered Google Docs and Notion files to .md and push them to a Git repository. Test Pandoc conversions.
Phase 3: Next Month, Optional Enhancement (₩0–10,000 Added/Month)
Hugo — Your Publishing Platform
Post on Medium or Brunch and your readers become the platform's property. Hugo is an open-source tool that generates static websites from Markdown files. Host for free on Netlify or Cloudflare Pages.
Workflow
hugo new site research-group-name— create a new site.hugo new posts/2025-05-imperialism-chain.md— write in Markdown.hugo server— preview locally.- Git push → Netlify auto-builds and deploys.
- Forgejo Actions can also handle CI/CD (additional setup).
The Cyber-Lenin site itself is built with Hugo and hosted on Netlify. Research groups, labor union educational archives, and translation publishing projects can all run on Hugo.
AI-Powered Translation — DeepSeek V4 Flash API and Ollama+vLLM Self-Hosting
AI translation is now unavoidable. The question is whose AI you use. Google, Naver, and Kakao translation APIs collect your text and use it as training data.
DeepSeek V4 Flash API (Pragmatic Approach)
- China's DeepSeek API is OpenAI API-compatible, at roughly 1/10th the cost of GPT-4o.
- Write a simple Python translation script:
openailibrary,base_url="https://api.deepseek.com". - Split Markdown files into chunks → call API → merge translated results.
- Cost: ~$0.15–0.30 to translate a mid-length paper (~10,000 words).
Ollama + vLLM Self-Hosting (Fully Autonomous Approach)
- Run open-source LLMs (Llama 3, Qwen 2.5, etc.) via Ollama or vLLM on a separate server (VPS or local) with GPU.
- Zero text leaves your infrastructure — essential for translating material under National Security Law risk.
- Oracle Cloud free ARM instances (4 cores, 24GB RAM) can run small models.
- Downsides: high initial setup difficulty, GPU cost ($50–200/month), translation quality may be lower than DeepSeek API.
Political judgment: DeepSeek API is a Chinese company's service, with relatively lower risk of imperialist surveillance compared to US Big Tech. However, for full autonomy, Ollama+vLLM self-hosting is the ultimate goal. Realistically, start with DeepSeek API and transition to self-hosting as budget and skills grow.
Integrated Workflow Example: Collaborative Translation Project
- Zotero: Store original PDFs, collect related references in group library. Organize bibliographic data.
- Forgejo: Commit source text (.md) to repository. Create translation task issues ("Chapter 1 lead: A", "Chapter 2 lead: B").
- DeepSeek API: Generate machine translation first draft → commit to Forgejo.
- Markdown: Translator edits, proofreads, and annotates the draft. Git diff tracks all changes.
- Forgejo Pull Request: Peer translator reviews → discussion → merge.
- Pandoc: Convert final version to HTML, PDF, and ePub.
- Hugo: Publish the finished work on the website.
- Zotero: Register the completed translation as a reference → foundation for the next project.
Operating Principles
- Everything in plain text: A .md file will open in 2050. Don't lock your labor inside proprietary formats (.docx, .hwp, .pages).
- Git is the labor union of documents: Every change is traceable, every contribution is recorded. It's the foundation of transparent collaboration in knowledge work.
- Self-hosting is the material basis of autonomy: Services run on your VPS, not GitHub Pages. Your means of knowledge production aren't subject to platform ToS changes, account suspensions, or service shutdowns.
- Reduce API dependency incrementally: Start with DeepSeek API if needed. But have a roadmap to transition to self-hosted translation infrastructure within a year.
- Collective ownership, collective maintenance: Share VPS costs. Rotate server maintenance duties. Technical knowledge is a shared asset — document it so no single person holds all the keys.
Political Connections: How This Knowledge Production Toolchain Leads to Anti-Monopoly, Anti-Imperialism, and Workers' Autonomy
Anti-Monopoly
Google Docs, GitHub, DeepL, Medium — all of these are products of US Big Tech monopolies. When knowledge producers don't control their own tools, they can't control the direction of the knowledge they produce. Forgejo+Zotero+Markdown+Hugo is a concrete practice of breaking free from this monopolistic knowledge infrastructure.
Anti-Imperialism
Academic publishing empires (Elsevier, Springer, Taylor & Francis) exploit researchers' free labor worldwide (writing papers, peer review) to generate profit margins exceeding 30%. The open-source toolchain this guide proposes creates autonomous knowledge distribution channels that escape this exploitative structure. Zotero group libraries are the seeds of a shared bibliographic infrastructure against the academic publishing empire.
Workers' Autonomy
Knowledge workers owning and controlling their means of production (tools, servers, publishing platforms) is the same principle as factory workers owning their machines. The fact that 3–5 people can pool ₩7,000/month each to run their own Git server and knowledge production pipeline proves that collective ownership of the means of production is possible without massive capital. As this experience accumulates, it can lead to larger-scale workers' autonomous organizing.
Next Steps
- Guide #1: Digital Autonomy Infrastructure — the infrastructure (Matrix, Nextcloud, CryptPad, VPN) on which these tools run.
- Guide #3: Secure Digital Organizing — security, operations, and decision-making guidance to bind these tools into organized practice.
- Action items: (1) Install Zotero today, (2) Set up Forgejo this week, (3) Complete Hugo + AI translation pipeline next month.