EverMind has open-sourced Raven, an agent runtime built on its EverOS platform. It's still in early pre-alpha, but Raven aims to solve one of the biggest headaches in AI agents: cross-session memory loss. By keeping user memory, agent memory, and world knowledge independently persistent, it promises long-term cognitive continuity.
Under the hood, Raven uses a built-in tool called SkillForge to automatically extract repeatable workflows from chat history and turn them into reusable local skills. It also introduces a Sentinel-based proactive trigger mechanism with Cron job support, so agents can send nudges and execute tasks asynchronously based on schedules or external events.
For ease of use, Raven offers both a command-line interface and a terminal UI (TUI), and ships with built-in gateways for 12 instant messaging platforms — including WeChat, WeCom (enterprise WeChat), Feishu (ByteDance's Lark), Slack, and Discord. With a single command, you can drop a memory-equipped AI assistant directly into your daily work chat.