A 100% local, private document search tool. It enables you to run deep, private searches across hundreds of pages per minute to get relevant context for your queries. Patense.local uses vLLM by default but can run on any backend LLM server using the OpenAI API, .
It basically breaks your references up into pages, passes each page to an LLM with the query, and asks if the content is relevant to the query. If it's relevant, it displays a short quote with a link to the full page.
- Deep document search - find relevant portions of references fast with AI
- Inventive feature extraction - get all disclosed features for possible claims or amendments
- NEW OA Auditor - Upload a set of claims and references, AI parses the claims into inventive elements and searches all references
- Privacy First: Run the tool entirely on your local machine, ensuring full control over your data.
- High Performance: Search and analyze large documents quickly and efficiently.
- Flexible Backend: While
text-gen-webuiis the default, Patense.local can work with any backend LLM server.
- vLLM (installation is outside the scope of this guide).
- Node.js and npm (These are necessary to run the application. If you're unfamiliar with installing them, it might be easier to use Patense.ai).
-
Clone the Repository
git clone https://github.com/JohnZolton/snorkle.git cd snorkle -
Install Dependencies
npm install
2.1 Rename .env.example to .env
-
Configure the Backend
Start your backend LLM server in api mode
in your text-gen-webui folder (or other backend) run:
vllm serve NousResearch/Meta-Llama-3.1-8B-Instruct --max-model-len 8000 --tensor-parallel-size 2 #Set to number of GPUs you're using -
Initialize the database in the /patense-local folder, run:
npm run db:push
-
Run the Application in the /patense-local folder, run:
npm run dev
-
Naviage to http://localhost:3000
Once the application is running, you can begin uploading documents and performing searches.