VCPToolBox

VCP (Variable & Command Protocol) is deployed between model APIs and front-end apps. Rather than a tool-calling framework, it aims to give an AI one continuous existence: a single timeline across web, mobile, desktop, group chats and email, so a conversation started on one client continues on another.

Its memory system, RiverMemo, treats memories as an associative network instead of plain vector similarity, and the heavy computation runs in a Rust native core. Before each request, VCP pre-computes what the model should know right now, such as relevant memories, time, weather, schedules and device state, instead of making the agent query for them. Agents can also schedule their own wake-ups, leave notes for later and work with other agents.

It runs as a Node.js server with Python dependencies or via Docker. The README warns that the agent gets low-level system permissions and asks users not to route it through unofficial relay APIs. The project is licensed CC BY-NC-SA 4.0, and its documentation is primarily in Chinese with English, Japanese and Russian versions.