Qalatra
A project manager your agent actually operates.
The pitch
AI agents can already read and update a lot of the tools teams use. The gap is the other direction: most project managers still treat AI like an outside assistant, not a worker you can assign, track, and fold back into the plan.
Qalatra closes that loop. Tasks, daily notes, habits, contexts, agent profiles, and dispatch history live in one local database, so work can move from "this needs doing" to "send this to an agent" without leaving the planning surface. The system can call into thousands of MCP servers, and whatever tools they expose, while keeping the task-management layer intact.
The work
The build
Electron keeps it native — Mac, Windows, and Linux from one codebase. SQLite lives on disk; nothing goes to a server. The MCP server runs locally via StreamableHTTP and can connect the app to the broader MCP ecosystem. The hard part was dispatch and synthesis — turning tasks, notes, habit history, and context into work packets an agent can act on without exhausting its context window. A built-in xterm.js terminal keeps agent work beside the task list. Open source under MIT.