Obsidian OCR

Extract text, formulas, tables, and structured content from images and PDFs directly into your Obsidian notes, powered by GLM-OCR.

This project is a fork of obsidian-latex-ocr by lucasvanmol. The original plugin focused exclusively on LaTeX formula recognition. Obsidian OCR extends this to full document OCR: any text, formula, table, or mixed content in an image can be extracted and inserted into your notes.

Features

• Full OCR — extract any text from images and PDFs, not just LaTeX formulas.
• Formula support — mathematical expressions are recognized and output in LaTeX.
• Paste from clipboard — use a custom command (e.g. Ctrl+Alt+V) to OCR an image from your clipboard and insert the result directly.
• Context menu — right-click any image in your vault and choose "Generate OCR text".
• Multiple backends — use the Hugging Face API for a zero-install cloud option, or run locally with Ollama or llama.cpp.
• On-demand local startup — local backends are started automatically when the first OCR query is made.

---

Using the Hugging Face API

The plugin can use the GLM-OCR model via the Hugging Face Inference API (free tier).

Setup

1. Create an account or log in at huggingface.co.
2. Generate a read access token in your Hugging Face profile settings. Creating one dedicated to this plugin is recommended.
3. In Obsidian, open Settings → Obsidian OCR and paste the token into the API Key field.

Limitations

• The free Inference API may take a few seconds to provision the model on the first request. Subsequent requests are faster.
• Rate limits apply on the free tier. If you hit them, wait a moment and retry.
• For heavy usage, consider running the model locally with Ollama or llama.cpp (see below).

---

Run Locally with Ollama

You can run GLM-OCR entirely on your machine using Ollama. No internet connection or API key required after the initial model download.

Requirements

• Ollama installed and available in your PATH (or at a custom path you configure).
• A vision-capable GLM-OCR model available in your Ollama instance.

Installation

1. Install Ollama

Download and install from ollama.com/download, then verify:

ollama --version

2. Pull the GLM-OCR model

ollama pull glm-ocr

The model is approximately 1–2 GB. The download happens once and is cached locally.

You can verify it is available with:

ollama list

3. Configure the plugin

In Obsidian, open Settings → Obsidian OCR:

• Enable Use local model.
• Set Local backend to Ollama.
• Set Ollama command/path — usually just ollama if it is in your PATH, or the full path to the binary.
• Set Ollama host — default is http://127.0.0.1.
• Set Ollama port — default is 11434.
• Set Ollama model — enter the model name exactly as shown by ollama list, e.g. glm-ocr.
• Use (Re)start backend, Check status, and Stop server when needed.

If Ollama is not reachable, the plugin tries to start it automatically on the first OCR operation.

---

Run Locally with llama.cpp

You can run OCR locally with llama.cpp using llama-server and a compatible model.

Requirements

• llama-server available in your PATH (or configured with a full path).
• A compatible OCR model, for example ggml-org/GLM-OCR-GGUF.

Example startup command

llama-server -hf ggml-org/GLM-OCR-GGUF --sleep-idle-seconds 300

Configure the plugin

In Obsidian, open Settings → Obsidian OCR:

• Enable Use local model.
• Set Local backend to llama.cpp.
• Set llama.cpp command/path — usually llama-server.
• Set Ollama host — typically http://127.0.0.1.
• Set Ollama port — typically 8080 (auto-set when selecting llama.cpp).
• Set llama.cpp startup args — default is -hf ggml-org/GLM-OCR-GGUF --sleep-idle-seconds 300.
• Use (Re)start backend, Check status, and Stop server when needed.

If llama.cpp is not reachable, the plugin tries to start it automatically on the first OCR operation.

VRAM note

--sleep-idle-seconds 300 can help reduce VRAM pressure during idle periods.

GPU Support (Ollama)

Ollama automatically uses your GPU if supported. To verify:

ollama run glm-ocr "test"

If you want to explicitly check CUDA availability, see the Ollama GPU documentation.

For llama.cpp GPU options, refer to the llama.cpp documentation and launch flags supported by your build.

Status Bar

The status bar at the bottom of Obsidian shows the current state of the backend:

Status	Meaning
OCR ✅	Ready
OCR ⚙️	Loading / warming up
OCR 🌐	Model being provisioned (API)
OCR 🔧	Needs configuration
OCR ❌	Unreachable

---

File Input Notes

• The ribbon modal supports selecting files and shows the selected filename.
• Image preview is shown for supported image formats; PDF preview is not rendered in the modal.

---

Attribution

• Forked from obsidian-latex-ocr by lucasvanmol.
• OCR powered by GLM-OCR by zai-org.