Allow dynamic referencing of local files in chats and projects

Projects are, by nature, living, evolving entities. Documentation grows, code changes, and scope transforms. This is the natural state of any meaningful project.

Despite this reality, even frontends with robust project grouping features (like claude.ai) treat knowledge as immutable artifacts requiring manual upload-delete-reupload cycles whenever local files change. For users with Obsidian vaults or similar knowledge bases, this process quickly becomes more time consuming than simply pasting file contents into the chat—partially defeating the purpose of structured context management.

While planned features like text file attatchments and project grouping with shared context will help T3 Chat achieve feature parity with mainstream AI offerings, I think T3 Chat has a real opportunity here to go even further by tackling this fundemental UX issue that the competition has yet to address.

The proposed solution

I propose a lightweight, browser-native approach on top of the previously mentioned planned features that does the following:

1. Enables filesystem access: Allow users to grant T3 Chat permission to specific files or user-defined local directories (e.g., an Obsidian vault) on their local filesystem

2. Implements dynamic referencing: When users type @ in chat, present a searchable dropdown of available local documents

3. Provides just-in-time content loading: When a file is selected or referenced via @filename, read its current contents directly from disk and attatch it to chat

4. Integration with projects: Allow users to define “default” knowledge on a per-project basis that is automatically attatched from the filesystem to new chats within projects (once they are implemented)

T3 Chat also isn’t an IDE like Cursor or an advanced RAG solution, so this hypothetical functionality would obviously need to be restricted to a limited amount of documents of reasonable size. After all, I don’t want to be able to dump my entire git repo into chats, just a modest amount of simple markdown documents.

Most importantly, T3 Chat isn’t a personal knowledge management system and shouldn’t try to be one. In my opinion, this approach respects that.

Benefits

• Users wouldn’t need to manually re-upload documents as they change

• It reduces the burden of T3 Chat needing to store files on its servers

• It would seamlessly integrate with other planned features like projects

• It could just leverage standard browser APIs

• Arguably better for privacy, I suppose

• Would be able to officially say that T3 Chat is superior to Claude’s frontend