MemoryBear v0.3.11 Community Release Notes — Where All Streams Converge

[keeees]

Release Date: July 8, 2026 | Codename: GuiCang (归藏 · Convergent Repository)

Building on v0.3.10's precision and transparency, v0.3.11 turns unification into a design principle. Memory configuration converges from the user level to the space level, memory and user routes align with the v1 standard, working memory admits multi-entry writes from API and MCP, and messaging behaviors harmonize across Agents and workflows — while cross-session context, cross-knowledge-base retrieval, and smarter document parsing lift what the platform can carry.

---

🚀 I. Core Upgrade Overview

1. Memory Intelligence 🧠

• Working Memory Multi-Entry Write: Working memory now accepts writes from both API and MCP entry points, with a unified surface for viewing working-memory content — one place to see everything, regardless of how it got in.
• Space-Level Memory Configuration: Memory configuration is elevated from the user level to the space level — one configuration governs the entire space, centralizing management and eliminating per-user drift.
• User Memory Library Deletion: A new operation lets you delete a user's memory library outright, cleaning up data that is no longer needed.

---

2. Knowledge Base & Document Processing 📚

• Smart DOC / PDF Parsing: DOC and PDF documents are now handled by an upgraded parser that preferentially produces Markdown and auto-chunks output, with strengthened image content recognition — a direct lift for knowledge retrieval accuracy.
• Cross-Knowledge-Base Retrieval: Retrieval and recall can now execute independently per knowledge base, compatible with heterogeneous models and retrieval configurations — cross-KB queries become both more accurate and more flexible.

---

3. Agent & Workflow 🔗

• Cross-Session Context: Agents and workflows gain cross-session context — recent conversation history is automatically referenced, so what was said before stays available now.
• Space-Level Model Configuration: Spaces now carry their own model configuration — pick the default model or a custom one, matched to each business scenario.
• Unified Memory Config in Extraction/Storage Nodes: Workflow "memory extraction" and "memory storage" nodes now share the space-level memory configuration, so configuration complexity drops and management stays consistent end to end.

---

4. API & Integration 🔌

• V1-Aligned Memory & User Routes: Server-side (api) memory-management and user-info routes are standardized to align with the V1 endpoint, giving internal and external interfaces a single, consistent shape.
• App Open v1 API: The application layer exposes a v1 API covering file upload, message feedback, suggested questions, conversation variables, app info, and annotation management — turnkey building blocks for integration and second-development.

---

5. Performance & Stability ⚡

• Summary & Insight Cache Refresh Optimization: The memory summary and insight cache-refresh mechanism is tuned for large user-base scenarios, lifting task execution stability and cutting timeout risk.
• Memory Storage Architecture Refinement: The memory storage module's underlying structure is refactored and refined, delivering a cleaner foundation and steadier system behavior.
• Sandbox Concurrency & Observability: The Sandbox environment gains stronger concurrent processing and more comprehensive logging, so runtime issues surface earlier and remedies land faster.

---

6. Interaction & Experience 🎨

• Unified Agent & Workflow Message Interactions: Message actions are harmonized across Agents and workflows — like, dislike, regenerate, favorite, copy, delete, and report are now available uniformly wherever a conversation lives.
• Documentation Center: The help center is redesigned as a full documentation center, making lookup and reading noticeably smoother.

---

🧭 Looking Ahead

v0.3.11 elevates convergence into a design principle. Memory configuration lifts from the user level to the space level, working memory admits API and MCP together, message interactions harmonize across Agents and workflows, and API routes align to a single v1 standard — the platform stops carrying multiple grammars and speaks one language. Alongside cross-knowledge-base retrieval, cross-session context, and smarter DOC / PDF parsing, this release begins to look less like a collection of features and more like a coherent surface to build on.

The forward-looking thread is depth of memory reasoning and control over its lifecycle. Cross-session context and unified memory writes prepare the ground for richer working memory and for a memory system that knows what to keep, what to weight, and what to let go. In parallel, the newly unified message-interaction surface and space-level configuration together give teams the levers they need to shape behavior at space granularity — not one user at a time.

Next, we will deepen the reflection pipeline's resilience and self-recovery, surface the memory-value weighting and cross-modal associative engines directly in the memory configuration experience, and progress a coherent forgetting mechanism aligned with capacity awareness. On the platform side, expect continued expansion of Agent memory tooling, broader knowledge-processing coverage, and a more expressive workflow authoring surface — with steady lifts to evaluation rigor and integration breadth throughout.

---

MemoryBear v0.3.11 社区版 发布说明 —— 万流归宗

发布日期： 2026年7月8日 ｜ 版本代号： 归藏（GuiCang · Convergent Repository）

在 v0.3.10 精准与透明的基础上，v0.3.11 将「归一」立为设计主线。记忆配置由用户级收敛为空间级，记忆与用户接口对齐 v1 标准，工作记忆开放 API 与 MCP 多入口写入，Agent 与工作流的消息交互全面统一；跨会话上下文、跨知识库检索与更聪明的文档解析共同抬升平台承载力。

---

🚀 一、核心升级概览

1. 记忆智能 🧠

• 工作记忆多入口写入：工作记忆同时支持 API 与 MCP 多入口写入，并统一记忆内容展示，多来源写入、一处可查。
• 记忆配置空间化：记忆配置由用户级调整为空间级，空间内统一配置与集中管理，告别按用户分散设定。
• 用户记忆库删除：新增用户记忆库删除能力，一键清理不再需要的用户记忆数据。

---

2. 知识库与文档处理 📚

• DOC / PDF 智能解析：DOC 与 PDF 文档解析升级，优先使用新版解析引擎生成 Markdown 并自动切块，同时增强图片内容识别，直接提升知识检索准确率。
• 知识库跨库检索：支持按知识库独立执行检索与召回，兼容不同模型与检索配置，跨库检索更精准、更灵活。

---

3. Agent 与工作流 🔗

• 跨会话上下文：Agent 与工作流新增跨会话上下文能力，自动引用最近历史消息，让此前的对话继续为当前服务。
• 空间级模型配置：空间支持模型配置，可选择默认模型或自定义模型，按业务场景灵活配置。
• 工作流记忆节点配置统一：工作流中的记忆提取与记忆存储节点统一使用空间级记忆配置，降低配置复杂度，管理更一致。

---

4. API 与集成 🔌

• 记忆与用户接口对齐 V1：服务端 api 记忆管理与用户信息相关接口路由与 V1 端保持一致，对内对外接口更规范清晰。
• 应用开放 v1 API：应用层开放 v1 API，覆盖文件上传、消息反馈、建议问题、对话变量、应用信息与标注管理等能力，便于系统集成与二次开发。

---

5. 性能与稳定性 ⚡

• 摘要与洞察缓存刷新优化：优化记忆摘要与洞察的缓存刷新机制，大规模用户场景下任务执行更稳定、超时风险显著降低。
• 记忆存储架构精修：优化记忆存储模块底层结构，为系统稳定性打下更整洁的底座。
• Sandbox 并发与可观测性：优化 Sandbox 并发处理能力与日志可观测性，运行更稳、问题更好定位。

---

6. 交互与体验 🎨

• Agent 与工作流消息交互统一：完善 Agent 与工作流的消息交互能力，点赞、点踩、重新生成、收藏、复制、删除、举报等操作全面统一。
• 文档中心焕新：帮助中心升级为全新文档中心，文档查阅更便捷、结构更清晰。

---

🧭 未来展望

v0.3.11 将「归一」升格为设计原则。记忆配置由用户级收敛为空间级，工作记忆同时接纳 API 与 MCP，Agent 与工作流的消息交互全面对齐，接口路由归入统一的 v1 标准——平台不再同时讲多种语法，只讲一种语言。配合跨知识库检索、跨会话上下文与更聪明的 DOC / PDF 解析，本次发布不再只是功能的堆叠，更像是一张可以在上面继续搭建的整齐平面。

本版本的前瞻主线是记忆推理的深度与记忆生命周期的可控。跨会话上下文与统一的记忆写入为更充沛的工作记忆铺路，也为一个懂得「留什么、重视什么、放下什么」的记忆系统蓄力。与之呼应，统一后的消息交互层与空间级配置一起，为团队提供了在空间粒度上塑造行为的杠杆——不必再逐个用户地调优。

接下来，我们将深化反思流水线的稳健性与自我恢复能力，把记忆价值权重与跨模态关联联想引擎带入记忆配置界面，并推进一套与容量感知联动的连贯遗忘机制。平台侧将持续拓展 Agent 记忆工具、扩宽知识处理覆盖，并带来更具表达力的工作流编排——同时在评测严谨度与集成广度上稳步抬升。

---

This discussion was created from the release MemoryBear v0.3.11 Community Release Notes — Where All Streams Converge.