DaiDocs Offers Local AI Memory Capture via Browser Extension

Overview

A new open-source project called DaiDocs is emerging as a potential solution to the growing concern over data sovereignty in the age of artificial intelligence. The project offers a browser extension and a local companion server designed to give users granular control over the digital footprints they create online.

By saving chosen web pages and chat interactions directly to local files, DaiDocs allows individuals to build personal knowledge bases that can be recalled by any compatible AI model later. This approach contrasts sharply with the current dominant paradigm, where user data is often harvested by cloud-based platforms and used to train opaque, centralized models without explicit user consent or ownership.

The project positions itself as a tool for personal memory augmentation, emphasizing privacy, local storage, and encryption. It aims to bridge the gap between fleeting online consumption and durable, personal AI recall, all while keeping sensitive data off remote servers.

Technical Architecture and Functionality

DaiDocs is built around two primary components: a pure browser extension and a lightweight local server. The extension itself is written in vanilla JavaScript and includes content scripts, a service worker, and dedicated pages for consent and privacy settings. It is designed to run on both Chromium-based browsers like Chrome and Edge, as well as Firefox.

The capture_server.mjs acts as the backend companion. It listens exclusively on the loopback interface at 127.0.0.1, ensuring that no data is transmitted over the internet. When the user selects content via the extension, the extension passes this data to the local server, which then writes it to plain text files on the user's machine.

Capture is off by default, never touches sensitive sites, and scrubs card numbers and national IDs before storage.

This architecture ensures that the user retains full ownership of their data. The files generated are simple, human-readable formats that can be ingested by local AI assistants, RAG (Retrieval-Augmented Generation) systems, or personal knowledge management tools. The project also includes a vault.mjs tool, which provides an encrypted-vault command-line interface for users who wish to secure their captured memories further.

The repository is structured to be self-contained, with a lib/ folder housing essential modules for token counting, secret redaction, session markers, versioning, and encryption. A dashboard tool is also included, which builds a visual "memory map" that the extension can open locally, giving users a graphical overview of their captured content.

Privacy Protections and Ethical Guardrails

Privacy is a central tenet of the DaiDocs project, as reflected in its comprehensive documentation including PRIVACY.md and ACCEPTABLE-USE.md. The developers have implemented strict guardrails to prevent accidental exposure of sensitive information.

Site Exclusions: The extension is programmed to never capture data from sensitive categories of websites. These include banking institutions, health portals, webmail services, and password managers. This proactive filtering reduces the risk of users inadvertently storing credentials or private health information in their local archives.

Data Scrubbing: Beyond site exclusion, the system employs active redaction techniques. Before any content is written to disk, the capture server scrubs potential personally identifiable information (PII). This includes credit card numbers and national identification numbers, ensuring that even if a device is compromised, specific financial or legal identifiers are not present in the plain text files.

No Keylogging: A critical privacy feature is the assurance that the extension does not capture what the user types. It captures rendered page content and chosen chat interactions, but it does not intercept keystrokes. This distinction is vital for maintaining trust, as it separates content consumption from input monitoring.

The project is governed by a clear license and detailed terms of use, emphasizing that it is intended for personal use only. The developers have also published a compliance checklist, indicating a serious approach to ethical software development and regulatory awareness.

Installation and User Experience

Getting started with DaiDocs is designed to be straightforward, requiring no complex package manager installations such as npm install for the basic operation. Users can launch the capture server by running the capture_server.mjs script directly with Node.js or by double-clicking a provided Windows launcher batch file.

To activate the browser component, users must load the extension in developer mode. This involves navigating to chrome://extensions or edge://extensions, enabling Developer mode, and selecting the extension/ folder. Upon first launch, the extension presents a one-time consent screen, after which users can manage capture settings through an on-page pill interface.

Workflow:

1. Start the local capture server.
2. Load the unpacked extension in the browser.
3. Accept the consent prompt.
4. Use the on-page pill to toggle capture for specific sites.
5. Access the memory dashboard for review.

This low-friction onboarding lowers the barrier to entry for non-technical users who wish to experiment with local AI memory. However, the existence of detailed technical documentation suggests that power users can further customize the vault encryption and integration with other AI tools.

Implications for the AI Ecosystem

The rise of projects like DaiDocs highlights a shifting sentiment in the tech community regarding data privacy and AI training. As generative AI becomes more ubiquitous, concerns about data scraping and lack of user control have intensified. DaiDocs offers a practical, decentralized alternative that empowers users to curate their own training data or personal context.

By enabling users to capture and encrypt their own web history and chat logs, the project supports the emergence of personalized AI assistants that are faithful to the user's specific context rather than general internet corpora. This could lead to more accurate, relevant, and private interactions with AI models.

Furthermore, the open-source nature of the project invites scrutiny and community contribution, which can enhance its security and reliability over time. It stands as a testament to the growing movement toward user-centric AI tools that prioritize transparency and local control.

As the landscape of AI continues to evolve, solutions that address privacy concerns through technical design rather than just policy promises may become increasingly important. DaiDocs represents a tangible step in that direction, offering a blueprint for how individuals can reclaim ownership of their digital memories in an AI-driven world.