LLMConnect

Inspiration

AI-driven conversations happen across many browser tabs and API accounts. We wanted a local-first workspace that keeps sessions isolated, preserves structured chat history, and makes it easy to reuse and organize prompt-driven work without sending everything to a third-party server.

What it does

• Runs multiple browser sessions in an integrated desktop dashboard.
• Extracts and stores structured chat history in a local SQLite database.
• Encrypts and manages API profiles locally so credentials stay on-device.
• Provides lightweight browser tooling and persona prompts to standardize interactions.
• Enables categorization and project tagging for later review and reuse.

How we built it

We built LLMConnect as a PyQt6 desktop application with a modular engine layer:

• UI: PyQt6 for the main window, dashboard, and browser tab management.
• Browser integration: in-process browser views with injected lightweight scripts for extraction and persona tooling.
• Storage: SQLite for chat history and a small encrypted store for API profiles.
• Modules: modular provider adapters so new LLM providers can be added by implementing provider interfaces in models/.
• Security: local-first architecture and encrypted credential storage to keep sensitive data on-device.

Challenges we ran into

• Keeping browser extraction robust against provider DOM changes.
• Managing isolated sessions while providing a fluid UX for switching contexts.
• Designing a small, usable encrypted profile store that is easy to configure.
• Balancing modularity with simplicity so new providers are easy to add.

Accomplishments we're proud of

• A working local-first desktop dashboard that brings browser chats into one place.
• Reliable structured extraction that makes conversations searchable and reusable.
• Secure, encrypted storage for API profiles and credentials.
• A modular design that lets contributors add provider adapters and custom skills easily.

What we learned

• Browser DOMs are a moving target — defensive extraction and graceful degradation are essential.
• Local-first privacy matters: users appreciate tooling that keeps keys and sessions on-device.
• Modular, well-documented interfaces dramatically lower the bar for community contributions.

What's next for LLMConnect

• Add more provider adapters and make integrations pluggable via a clear plugin API.
• Improve extraction resilience with heuristics and optional user-assisted selectors.
• Add sync options for users who want encrypted cloud backups.
• Build example workflows and templates for common roles (developer, researcher, product manager).

Built With

• anthropic
• css
• google-gemini
• html5
• openai
• pyqt6
• pyqt6-webengine
• python
• sqlite