Docux

Less is More

[Docux @docuxstd]

I'm working on a project using Claude Opus right now, and I feel like it's gotten worse—it's just bug after bug; fixing one issue only reveals another. 🤬🤬🤬

[Docux @docuxstd]

New:
- Logo like a Turbin,
- New theme
- New OCR to Select/Copy on Scan PDF
Please remain free software forever, Torreader PDF.

[Docux @docuxstd]

I just discovered Toreader PDF, open-source.
🚀 Open 200MB PDFs in 1–2s.
📑 Merge PDFs without losing bookmarks, split/extract/insert pages
🌍 Translate PDF
✍️ Sign PDFs
💾 Safe editing
Look at https://t.co/ip1HG8wHBS
#PDFreader #opensource #MIT

[Docux @docuxstd]

@FelixNgHuy I found you, goodjobs

[Docux @docuxstd]

I'm following ToreaderPDF and this version 2.2.4... the app is still fast and smooth, but this version zooms in so clearly you can see details down to 0.1mm, micro-text, it has OCR search, handwriting recognition, and highlighting. Is it really free?
#PDFreader #opensource #MIT

[Docux @docuxstd]

Adobe Photoshop -> GIMP
Adobe Illustrator -> Inkscape
Sketchup -> Blender
and???

[Docux @docuxstd]

@artic_ai I found this Compares document on a PDF Toreader website, but I'm not sure if it's correct.

[Docux @docuxstd]

I'm a bit skeptical about this Toreader PDF repository. It can open a 950MB, 300-page PDF drawing file in just 10 seconds, and it still loads normally even at page 200 or page 50. It only uses 600MB of RAM. This 2.2.3 version is completely different from the previous garbage.

[Docux @docuxstd]

I just discovered Toreader PDF, open-source.
🚀 Open 200MB PDFs in 1–2s.
📑 Merge PDFs without losing bookmarks, split/extract/insert pages
🌍 Translate PDF
✍️ Sign PDFs
💾 Safe editing
Look at https://t.co/ip1HG8wHBS
#PDFreader #opensource #MIT

[Docux @docuxstd]

@49agents I reported the error, Claude Code scanned my codebase for errors, dispatched OpenCode Deepseek v4 Flash to fix the code, but the quota was still used up too quickly.

[Docux @docuxstd]

When you urgently need to fix a bug - and Claude Code...

[Docux @docuxstd]

I just discovered Toreader PDF, open-source.
🚀 Open 200MB PDFs in 1–2s.
📑 Merge PDFs without losing bookmarks, split/extract/insert pages
🌍 Translate PDF
✍️ Sign PDFs
💾 Safe editing
Look at https://t.co/ip1HG8wHBS
#PDFreader #opensource #MIT

[Docux @docuxstd]

🤖 Private RAG chatbot — all open-source:

- LLM: Llama 3, Ollama
- Embed: nomic-embed, bge
- Vector DB: Qdrant, Chroma
- Framework: LlamaIndex, LangChain
- Parse: unstructured, pypdf
- API: FastAPI
- UI: Streamlit

Wiring in the diagram ↓

#AI #RAG #LLM

[Docux @docuxstd]

🌐 How to code from a browser and never lose a job?

- Editor: code-server, VS Code in the browser
- Reach it: Cloudflare Tunnel + Access, no ports
- Durable shells: tmux, detach and reattach
- Close the tab, the build keeps running

The wiring ↓

#RemoteDev #DevTools #AI #LLMs

[Docux @docuxstd]

🤖 Private RAG chatbot — all open-source:

- LLM: Llama 3, Ollama
- Embed: nomic-embed, bge
- Vector DB: Qdrant, Chroma
- Framework: LlamaIndex, LangChain
- Parse: unstructured, pypdf
- API: FastAPI
- UI: Streamlit

Wiring in the diagram ↓

#AI #RAG #LLM

[Docux @docuxstd]

🧮 Which model @Zai_org is actually cheapest to code?

Same 3 tasks, three tiers benched:

- 5.2: 2496 tokens, 43s, fewest and fastest
- 4.7: 3384 tokens, 133s, slowest
- 4.5-air: 3754 tokens, 93s, heaviest
- Twist: 5.2 fewest tokens but 2-3x quota

Read the tradeoff ↓

#AI #LLM

[Docux @docuxstd]

💳 API tokens or a flat coding plan, which is cheaper?

- API: pay per token, drains per call
- Plan: flat fee, billed by request quota
- Trap: a heavy model burns 2-3x quota
- Promo windows pull it back to 1x
- Off-plan model = $0 wallet stall

Pick by how you code ↓

#AI #LLM

[Docux @docuxstd]

💳 API tokens or a flat coding plan, which is cheaper?

- API: pay per token, drains per call
- Plan: flat fee, billed by request quota
- Trap: a heavy model burns 2-3x quota
- Promo windows pull it back to 1x
- Off-plan model = $0 wallet stall

Pick by how you code ↓

#AI #LLM

[Docux @docuxstd]

🔌 How to run Codex CLI on a non-OpenAI model?

- Codex dropped chat API, Responses only now
- Your provider speaks only chat completions
- Bridge: LiteLLM proxy, port 4037
- Prefix zai/ not openai/ to translate
- Hook: drop tool type custom or 400

The wiring ↓

#AI #DevTools

[Docux @docuxstd]

🔌 How to run Codex CLI on a non-OpenAI model?

- Codex dropped chat API, Responses only now
- Your provider speaks only chat completions
- Bridge: LiteLLM proxy, port 4037
- Prefix zai/ not openai/ to translate
- Hook: drop tool type custom or 400

The wiring ↓

#AI #DevTools

[Docux @docuxstd]

🆓 How to build an all-free LLM inference stack?

- Main coder: DeepSeek V4
- Research: Gemini, 1500 req/day free
- Fast inference: Groq, Cerebras, NVIDIA
- Offline fallback: Ollama (qwen)
- Pay for one thing: the orchestrator brain

The stack ↓

#AI #LLM

[Docux @docuxstd]

@LightningD11921 Use ChatGPT or Codex or Claude code, they will setup to you, bro

[Docux @docuxstd]

🍓 How to self-host an API on a Raspberry Pi?

- App: FastAPI or Node, kept alive by PM2
- Expose it with a Cloudflare Tunnel
- No port-forward, DNS + TLS free at the edge
- Works behind CGNAT, survives reboots

The wiring ↓

#SelfHosted #RaspberryPi

[Docux @docuxstd]

🆓 How to build an all-free LLM inference stack?

- Main coder: DeepSeek V4
- Research: Gemini, 1500 req/day free
- Fast inference: Groq, Cerebras, NVIDIA
- Offline fallback: Ollama (qwen)
- Pay for one thing: the orchestrator brain

The stack ↓

#AI #LLM

[Docux @docuxstd]

🧩 How to build a portable desktop PDF tool?

- Render + edit: PDFium (MIT, no AGPL trap)
- Structure + bookmarks: QPDF
- Digital signing: OpenSSL CMS, PAdES
- Translate: Gemini + a 2-tier cache
- UI: Qt, one self-contained exe

The stack ↓

#Cpp #PDF

[Docux @docuxstd]

🤖 Private RAG chatbot — all open-source:

- LLM: Llama 3, Ollama
- Embed: nomic-embed, bge
- Vector DB: Qdrant, Chroma
- Framework: LlamaIndex, LangChain
- Parse: unstructured, pypdf
- API: FastAPI
- UI: Streamlit

Wiring in the diagram ↓

#AI #RAG #LLM

[Docux @docuxstd]

✍️ How to digitally sign a PDF from code (PAdES)?

- Add a sig dict: adbe.pkcs7.detached
- Reserve a /ByteRange over the whole file
- Hash everything except the signature hole
- Sign it with OpenSSL CMS, then self-verify

The byte dance ↓

#PDF #Crypto

[Docux @docuxstd]

🆓 How to build an all-free LLM inference stack?

- Main coder: DeepSeek V4
- Research: Gemini, 1500 req/day free
- Fast inference: Groq, Cerebras, NVIDIA
- Offline fallback: Ollama (qwen)
- Pay for one thing: the orchestrator brain

The stack ↓

#AI #LLM

[Docux @docuxstd]

@LightningD11921 Why don't you buy a cheap domain name, use Cloudflare Tunnel to attach a subdomain, and you can access the Pi remotely?

[Docux @docuxstd]

🆓 How to build an all-free LLM inference stack?

- Main coder: DeepSeek V4
- Research: Gemini, 1500 req/day free
- Fast inference: Groq, Cerebras, NVIDIA
- Offline fallback: Ollama (qwen)
- Pay for one thing: the orchestrator brain

The stack ↓

#AI #LLM

[Docux @docuxstd]

🧠 How to run a full dev team solo with AI?

- You drive: an orchestrator model
- Coder: a cheap, strong model
- Reviewer: a DIFFERENT model, catches more
- Researcher: a search-grounded model
- Handyman: a free router for tiny edits

Steal the wiring ↓

#AI #BuildInPublic