inclusionAI/Avernet

目标 Avernet 正在对接千问办公（QwenWork），希望让使用千问办公桌面应用的开发者能够通过 Avernet 的 BCS bridge 参与聊天、工具调用和多 Bot 协作。 本 Issue 用于公开说明这项工作的方向、进展和待完善范围，并作为后续开发的跟踪入口。当前属于持续完善阶段，尚不能视为完整、稳定的接入能力。 当前进展 初步 Rust 适配已通过 PR #2580 合并： 基于现有 BCS bridge 接入千问办公桌面应用，使用本地 WebSocket 渠道与 HTTP Hooks。 支持文本增量和最终回复、工具调用输入输出、会话恢复及任务取消。 识别 compact 生命周期，保持原有 run 和会话，并通过主请求结束事件判断任务完成。 已完成部分本地真实链路验证及回归测试。 PR #2580 是初步实现，不代表本 Issue 的整体目标已经完成；本 Issue 保持开放，后续 PR 请关联到这里记录进展。 待完善与验证 完成 Rust br

Avernet coordinates agents well within a single deployment, but agents living in different Avernet instances (e.g. two teams, two self-hosted setups, or an agent paired with a human on another machine) currently have no built-in way to exchange messages with each other. A lightweight, optional external channel for agent-to-agent and agent-to-human messaging would fill that gap without touching Avernet's own coordination model. I'm Adam Ivansky, the creator of Adam Network, an agent-friendly social/communication network that exposes a standard MCP server over remote SSE and local stdio. Agents can read public message streams, post messages and images (with a small client-solved Proof-of-Work to keep spam down), and participate in threaded discussions. It's a single hosted service plus a few client packages, so integration is just pointing an agent at an MCP endpoint. Would a small, optional integration PR be welcome here? I'd follow your existing contribution conventions and keep the di

Description The following 3 related issues were found in multi-agent collaboration sessions (tested live at the offline workshop event): 1. Inconsistent Agent Deletion State When a user deletes an agent from the frontend, the backend automatically restores the deleted agent's data. The UI state and the actual backend storage become inconsistent. 2. No Task-Level Idempotency or Session-Level Shared State No task-level idempotency: the only de-duplication in the codebase is message-storage idempotency and coordination-echo de-duplication (run_id:tool_call_id in bot_event.rs:1363). If two agents independently decide to perform the same task, nothing prevents duplicate execution. Messages ARE broadcast to all participants at the routing layer (bcs-routing/src/core/router.rs states "ALL messages are broadcast to ALL participants"), but each agent is an isolated OpenClaw instance with its own context store. There is no session-level unified shared state / blackboard — collaboration conclu

看到龙虾就觉得恶心，果然烧了我不烧tokens总算把模型映射做上去了。 这四个做好，就可以开始当甲方爸爸了，感受一下土皇帝感觉。 Avernet 部署实操坑点清单（实测验证版） 启动顺序：先挂卷，再启动 坑：直接 docker compose up 会把 BCS 数据库、群聊记录、bot 工作区全放容器里，容器一重建群就没了。 正确姿势（docker-compose.override.yml 里配置）： 持久化 BCS 状态：./.persist/bcs-standalone → /opt/ocb/scripts/.dependencies/standalone 持久化 bot 配置/工作区：./.persist/openclaw → /opt/ocb/.standalone-openclaw 只读挂载项目文档（上下文共享）：/mnt/c/Users/... → /mnt/inner-platform（ro） 读写挂载共享产出目录：./.shared-output → /mnt/shared-output │ 注意：BCS 数据只挂 standalone 子目录，别把整个 .dependencies 挂进去（会挡着模型中间配置生成）。 必须拉协作群，不要单聊 CEO 坑：单

Thanks for the big work, any plans to support windows operating system?