Essai: Privacy-First AI Assistant for Academic Writing

Essai is a stateless, privacy-first academic writing assistant that provides structured evaluations for academic writing by dynamically constructing prompts and proxying to large language models. Maintained by JohnKing376, this project is open-sourced on GitHub (link: https://github.com/JohnKing376/essai-backend) and was released on June 5, 2026. It aims to address the privacy concerns (storing user data) and lack of structured academic norms in existing AI writing tools.

With the rapid development of Large Language Model (LLM) technology, the field of academic writing is undergoing profound changes. However, existing AI writing tools have two core issues: first, they need to store user data, which raises privacy concerns; second, their output lacks structured academic norms. The Essai project is a solution designed specifically for these pain points.

Essai is a backend service system designed specifically for academic scenarios. Its core feature is the adoption of a stateless architecture—meaning the system does not persistently store any user input content or generated results. All processing is completed within the request cycle, after which resources are released, fundamentally eliminating the risk of data leakage.

Stateless Design Philosophy

The stateless architecture makes each request independent, and the server does not retain user context, bringing three major benefits: maximum privacy protection, easy horizontal scaling, and simple fault recovery.

Dynamic Prompt Construction

Essai does not use fixed templates; instead, it intelligently assembles prompts based on the user's input text type, subject area, and needs, ensuring that different disciplines receive targeted suggestions.

Multi-Document Format Support

It supports processing raw text and documents in formats like PDF and Word, automatically extracting content for analysis and lowering the user's entry barrier.

Essai achieves its privacy-first approach through three mechanisms:

1. Zero Data Persistence: All processing is done in memory and not written to disk;
2. Proxy Mode: Acts only as a transparent proxy between the user and the LLM, without retaining intermediate results;
3. Structured Output: The returned writing evaluations are presented in a structured format for easy user understanding and application.

Essai is suitable for various academic writing scenarios:

• First draft evaluation: Quickly identify logical flaws, insufficient arguments, and other issues;
• Language polishing suggestions: Optimize expression to meet academic norms;
• Structural integrity check: Ensure all parts of the paper comply with academic standards;
• Citation format assistance: Check the consistency of citation formats.

For college students, researchers, and academic editors, it is a powerful and secure writing assistant tool.

Limitations

• The stateless design cannot provide cross-session personalized suggestions;
• Relies on external LLM APIs, limited by model capabilities and availability;
• Requires users to deploy it themselves, which has certain technical skill requirements.

Future Directions

• Add local model support to reduce reliance on external APIs;
• Develop a frontend interface to improve user experience;
• Explore more advanced protection mechanisms like differential privacy.

Essai represents a new design paradigm for AI assistant tools—while providing intelligent services, it prioritizes user privacy protection. In today's era where academic integrity and data privacy are increasingly valued, this concept has important practical significance. For academic workers concerned about data security, Essai is an open-source project worth paying attention to.