PilotDeck
An open-source, task-oriented AI agent productivity platform organized around per-project workspaces, with traceable long-term memory, per-task cost tracking and automatic routing between models.
PilotDeck is an agent operating system built around the concept of a WorkSpace, aimed at long-running, multi-project productivity work rather than one-shot programming or single-turn question answering. Each WorkSpace isolates its own files, memory and skills from other projects.
It focuses on three capabilities: traceable memory that lets a user see and edit which memory entry led to an agent mistake without restarting a conversation; routing that matches tasks of different difficulty to different models instead of always using the most capable one; and an always-on mode that lets an agent discover work, report progress and save results as files while the user is away. It supports the Model Context Protocol and behaves consistently across web, CLI and instant-messaging front ends.
It was jointly developed and open-sourced by Tsinghua University's THUNLP group, ModelBest, OpenBMB and AI9Stars.