Loading content...
Conversation-first project management where an assistant and a lightweight dashboard share one authoritative workspace.
A GPT-native project-management alpha where authenticated MCP tools and a visual dashboard operate on the same authoritative work records.
Users increasingly plan work inside AI assistants, but the useful output often remains trapped in a conversation. Copying it into a separate project-management tool recreates the work. Docket explores a different model: the assistant becomes an operating interface for projects and tasks, while a lightweight dashboard preserves visual inspection and direct control. The product must prevent a model from silently overwriting newer work and must avoid becoming another system that demands access to every customer tool.
The private alpha stores user-owned projects and tasks in Cloudflare D1 through a typed application layer. The dashboard supports board and list views with editable title, description, status, priority, owner label, due date and source link. Authenticated MCP tools expose narrow operations for listing work and creating or updating records. Both surfaces use the same backend. Optimistic version checks reject stale updates rather than letting old conversational context overwrite newer changes. External files and services remain links or user-managed assistant connections; Docket does not copy data it does not need.
Keep authoritative project records in Docket while users manage external source connections in their chosen assistant.
Expose narrow authenticated MCP tools rather than giving the model direct database access.
Use optimistic task versions to prevent silent overwrites from stale conversational context.
Let conversational and visual interfaces operate on the same records.
Use managed authentication and Cloudflare D1 for the hosted alpha while clearly documenting features that remain unimplemented.
Test the RAG retrieval pipeline by asking specific questions about this case study: