What is docket?

Docket is a library for TypeScript apps that lets an AI agent search through years of business email and attachments and get answers that trace back to the exact original message. It runs fully inside your own application using SQLite for storage, so there is no separate server or cloud service to set up. The project's stance is that similar tools, like RAGFlow or R2R, are Python services meant to run alongside your app, which is a heavy dependency if your app is written in TypeScript and your data is just one company's mail archive. Docket instead installs like any other library and stores everything in a single directory on disk. Its design keeps the original email and attachment files untouched and permanent, treating anything derived from them, such as search indexes or AI embeddings, as something that can always be rebuilt. Every answer can point back to the exact bytes it came from. Rather than a single search that returns a fixed number of top results, an AI agent using Docket calls a set of tools to filter, search, follow a conversation thread, look at a timeline, or fetch a specific source. Search itself combines traditional keyword matching with vector based search, then reorders results with a reranking step. Docket also keeps a ledger of facts extracted from the mail, where each fact records when it became true in the real world and when the agent learned it, so newer information replaces older information without deleting the history, which lets you ask what was true as of a specific date. Email threads are reconstructed by matching message headers rather than by parsing subject lines. The package includes a command line tool that exposes these same features to any MCP compatible AI client, either read only or with permission to write new facts. It requires Node.js 20 or newer and is released under the MIT license.