BookLLM

BookLLM is a self-hosted book translation app with glossary, review, and polish pipelines.

It supports OpenAI-compatible APIs and local LLM runtimes for long-form reading and iterative translation workflows.

Current language options support translation between Chinese, English, Japanese, Korean, French, German, Spanish, and Russian. Source language can also be set to auto-detect.

_{Bilingual long-form reading with live translation output.}

Features

OpenAI-compatible Works with OpenAI-compatible providers and runtimes, including OpenAI, DeepSeek, OpenRouter, Ollama, LM Studio, and OMLX.	Live Reader Stream translations directly into a bilingual long-form reading interface.	Multi-stage Pipeline Translate with optional glossary extraction, review, and polishing stages.
Sidekick Model Use Sidekick as an optional second model for review, polishing, or quality-focused stages.	Glossary Support Maintain per-page glossary context, with optional whole-book extraction.	PDF / OCR Support Extract text from PDFs and use bundled OCR for scanned pages.

Screenshots

Library Books, progress, and translation tasks.

Translation Pipeline Glossary, review, and polishing configuration.

Pipeline Cost Notice

[!IMPORTANT]

Enabling optional pipeline stages can significantly increase token usage, API cost, and processing time.

These stages may improve translation quality, but the marginal gain is not always large enough to justify the extra cost and latency, especially for everyday translation tasks.

Before using the full pipeline on a long document, test it on a short file or a small excerpt first. Compare output quality, latency, and token usage, then enable only the stages that clearly help your use case.

Quick Start

Run BookLLM locally with Docker Compose. No .env file is required for single-machine use.

git clone https://github.com/purecodework/bookllm.git
cd BookLLM
docker compose up -d --build

Then open:

http://localhost:3000

For later starts, use:

docker compose up -d

Rebuild only after source, dependency, or Dockerfile changes:

docker compose up -d --build

Initial Setup

1. Open Model Connection from the sidebar.
2. Configure your OpenAI-compatible endpoint:
   • Base URL: an OpenAI-compatible /v1 endpoint, such as https://api.openai.com/v1
   • For a host-local runtime such as Ollama, http://localhost:11434/v1 is accepted; Docker deployments normalize it internally when needed
   • API key
   • Model
3. Optional: if you are using a hosted API provider instead of a local runtime, tune the input token budget and concurrency in Settings to improve throughput.
4. Upload an EPUB, PDF, or TXT file to begin translation.

Architecture

Next.js frontend
  -> NestJS API + SSE
     -> PostgreSQL: books, pages, settings, glossary/context, token usage
     -> Redis: BullMQ queues, job state, event fanout
     -> Translation workers: chunking, glossary/context, translation, review, polish, export
     -> FastAPI OCR service: PyMuPDF + Tesseract
     -> Local asset storage: originals, covers, extracted images
     -> OpenAI-compatible LLM endpoints: primary + optional Sidekick

Known Limitations

• Token usage shown in the UI is approximate. Provider dashboards remain the billing source of truth.
• OCR does not preserve the original layout.

License

MIT