DeepCode v2.2.0 makes long sessions dependable and puts the model under your control: sessions survive resume, compaction, and a second window; settings, models, and reasoning are declared rather than guessed; sub-agents and MCP servers start when needed and stop cleanly; and the Desktop speaks Simplified Chinese. It also folds in a month of community fixes and security hardening.

Highlights

• Sessions that hold up. Resuming a session restores what the agent did, not only what it said; compaction keeps the recent tail and can be triggered on demand with /compact; context pressure is measured from the provider's own usage; and two windows on one session no longer crash each other — one live writer per session, enforced.
• Your models, declared. A dsh-style settings dialog with per-model declarations (context window, capacities, reasoning ladder), live model discovery per connection, a Session-level context-window cap (/context 64k), and keys that reach exactly the requests they were resolved for.
• A terminal that reads like an app. Every bare command opens a selector, the TUI has a status line and tool cards, and a new turn no longer discards the prompt cache.
• Sub-agents and MCP, on demand. Delegate to Codex or Claude Code as sub-agents, hold follow-up conversations with native ones, activate MCP servers lazily, and let hooks see the whole lifecycle (SessionEnd, PreCompact).
• Memory with boundaries. Compaction leaves a note in workspace memory, and injected memory content sits inside an escaped <untrusted-data> boundary, so a poisoned note cannot forge instructions.
• Desktop in Simplified Chinese, eight themes including a high-contrast one, an appearance scale that actually resizes the app, and a project switch that no longer strands the MCP catalog.
• Hardening. Command-executor bypasses closed, native file operations routed through the sandbox, Windows private files protected by a fail-safe ACL, Plugin manifests warned about plaintext credentials, and dependency advisories (pip, pypdf, browserslist) cleared.

Thanks

This release includes contributions from @atirna, @lihuiyang1024, @rifkir23, @MarkHe1222, @imshunsuke, and @raymondginger2018-sudo — thank you.

Full changelog: v2.1.0...v2.2.0

DeepCode v2.2.0 让长会话变得可靠，并把模型真正交到你手里：Session 经得住恢复、压缩和第二个窗口；设置、模型与推理档位改为显式声明而非猜测；子代理与 MCP 服务器按需启动、干净退出；Desktop 说简体中文。同时合入了一个月的社区修复与安全加固。

亮点

• 站得住的 Session。 恢复会话时找回的是 Agent 做过什么而不只是说过什么；压缩保留最近的尾部，也可用 /compact 按需触发；上下文压力按服务商报告的真实用量测量；同一 Session 开两个窗口不再互相崩溃——每个 Session 只有一个活跃写入者，由系统锁保证。
• 你的模型，显式声明。 dsh 风格的设置对话框，逐模型声明（上下文窗口、容量、推理档位），每个连接可实时发现模型，Session 级上下文上限（/context 64k），密钥只到达为它解析的请求。
• 像应用而不是日志的终端。 每个裸命令都打开选择器，TUI 有状态行与工具卡片，新的 Turn 不再丢掉提示缓存。
• 按需的子代理与 MCP。 可把任务委派给 Codex 或 Claude Code 子代理，与原生子代理进行后续对话，MCP 服务器懒启动，Hooks 看得到完整生命周期（SessionEnd、PreCompact）。
• 有边界的记忆。 压缩会在工作区记忆里留下一条笔记；注入的记忆内容包在转义过的 <untrusted-data> 边界里，被投毒的笔记无法伪造指令。
• 简体中文的 Desktop，八套主题（含高对比度），真正缩放整个应用的字号偏好，以及切换项目不再让 MCP 目录卡住。
• 安全加固。 堵上命令执行器的绕过，原生文件操作走沙箱，Windows 私有文件获得 fail-safe 的 ACL，插件清单对明文凭据发出警告，并清除了 pip、pypdf、browserslist 的依赖公告。

致谢

本版本包含来自 @atirna、@lihuiyang1024、@rifkir23、@MarkHe1222、@imshunsuke 与 @raymondginger2018-sudo 的贡献——谢谢你们。

DeepCode v2.1.0 unifies Skills, Agent Plugins, and MCP across CLI, TUI, and Desktop. It includes reviewed upstream Skills, a managed MCP catalog with OAuth and real connection tests, portable local Plugins, and verified packaged-runtime resources.

What's Changed

• chore: land PRs #139/#148/#149 with original authorship by @Zongwei9888 in #156
• Revert Windows private-file ACL restriction (broke main) by @Zongwei9888 in #158
• docs: add CONTRIBUTORS.md by @Zongwei9888 in #157
• fix: import LoopDetector, ProgressTracker by @crsmillan in #119
• fix: surface detailed workflow errors in the UI by @yuefengw in #131
• fix(backend): prevent path traversal in SPA catch-all route (#126) by @AAtomical in #135
• Update codebase_index_workflow.py by @muhammadibrahim313 in #14
• Create registry.py by @User1234608 in #20
• Update mcp_tool_definitions_index1.py by @lindomar8 in #21
• Agregué una contribución desde mi fork by @PacoMortal in #23
• Fix: Command Injection Vulnerability in cleanup_cache by @Mirza-Samad-Ahmed-Baig in #27
• fix: delete the source file by @dingdinglz in #48
• Update .gitignore: Add logs/, deepcode_lab/papers/, and mcp_agent.sec… by @VinodHatti-AI-Developer in #49
• fix(security): allowlist the #131 test fixture in gitleaks history by @Zongwei9888 in #160
• feat(providers): add Requesty, MiniMax and Forge by @Zongwei9888 in #159
• fix(providers): resolve reasoning through the catalog; add appearance preferences by @Zongwei9888 in #161
• fix(providers): gate thinking on model capability; add an installed-font picker by @Zongwei9888 in #162
• docs: add the Trendshift ranking badge beside the trending one by @Zongwei9888 in #163
• fix(security): bump js-yaml past CVE-2026-59870 by @Zongwei9888 in #168
• fix(security): raise the aiohttp floor past the known advisories by @Zongwei9888 in #166
• docs: add 2026-08-07 news entry and an Appreciation section by @Zongwei9888 in #165
• fix(cli): install log sinks at startup so internals stay off the terminal by @Zongwei9888 in #169
• fix(ci): stabilize and scope desktop checks by @Zongwei9888 in #173

New Contributors

• @crsmillan made their first contribution in #119
• @yuefengw made their first contribution in #131
• @AAtomical made their first contribution in #135
• @muhammadibrahim313 made their first contribution in #14
• @User1234608 made their first contribution in #20
• @lindomar8 made their first contribution in #21
• @PacoMortal made their first contribution in #23
• @Mirza-Samad-Ahmed-Baig made their first contribution in #27
• @dingdinglz made their first contribution in #48
• @VinodHatti-AI-Developer made their first contribution in #49

Full Changelog: v2.0.0...v2.1.0

After continuous iteration, HKUDS DeepCode v2 is officially released 🚀
An open-source, general-purpose Coding Agent built for real-world software engineering.

Coding Agents are becoming increasingly capable of generating high-quality code. However, developers still face several recurring problems in real projects:

• ⏸️ Complex tasks stop halfway and require repeated prompts to continue
• 🧪 Code appears complete, but tests, builds, and runtime behavior are not actually verified
• 🔁 Regression checks, test repairs, documentation updates, and repository maintenance must be initiated manually each time

DeepCode v2 is designed to reduce this repeated communication and manual work.

Instead of only generating code, it helps developers continuously move tasks forward inside real repositories until the results are verified and ready for review.

What is new in DeepCode v2:

• 🔄 Loop Engineering
  DeepCode continuously understands, implements, verifies, and repairs work around a natural-language Goal. It does not stop after producing a single response, but keeps progressing based on real execution results.

• 🎛️ Control throughout execution
  Developers can add requirements, revise the Goal, pause, resume, or change direction while the Agent is running, without losing work that has already been completed.

• ✅ Verified delivery
  DeepCode actually runs commands, tests, and builds, and delivers results together with code changes and execution evidence, making them easier to review, validate, and merge.

• 🤖 Personalized Automation
  Developers can create project-specific tasks using natural language and run them manually or on a schedule for regression checks, test repairs, documentation synchronization, and repository maintenance.

• 💻 Unified CLI and Desktop experience
  The same project, Session, Goal, and execution history can be continued across both CLI and Desktop, reducing context loss when switching tools.

We hope DeepCode v2 helps open-source developers and engineering teams spend more time on meaningful design, decision-making, and creation—and less time repeatedly prompting Agents, rechecking results, or explaining the same task again.

Less prompting. More completed work.

Try DeepCode v2, share your feedback, open an Issue, or join us in building it together. 🚀

DeepCode v1.3.0 完成了 V2 架构升级，将 DeepCode 扩展为更统一、安全且可扩展的通用智能编程代理。

• 统一 CLI、Web 与 Headless 模式的 Agent 内核和事件协议，同时保留原有 Paper2Code 工作流。
• 新增交互式 CLI、原生编程工具、持久化 Session、项目记忆与长对话上下文管理。
• 引入权限控制、安全沙箱、工具审批及工作区隔离机制。
• 支持自主循环、定时任务、并行团队与模型驱动的子代理协作。
• 支持 Skills、生命周期 Hooks、分层配置，以及在任意目录中启动和恢复 DeepCode。

DeepCode v1.3.0 delivers the V2 architecture upgrade, evolving DeepCode into a more unified, secure, and extensible general-purpose coding agent.

• Unified the agent runtime and event protocol across CLI, Web, and headless modes while preserving the existing Paper2Code workflow.
• Added an interactive CLI, native coding tools, persistent sessions, project memory, and long-context management.
• Introduced permission controls, security sandboxing, tool approvals, and workspace isolation.
• Added autonomous loops, scheduled tasks, parallel teams, and model-driven subagent collaboration.
• Added Skills, lifecycle Hooks, layered configuration, and support for launching and resuming DeepCode from any directory.

本次更新主要包括：
DeepCode + nanobot 集成：支持通过 飞书机器人 调用 DeepCode，在聊天中发起论文转代码、对话转代码和任务进度查询，并新增飞书实战截图示例。
文档重构：全面更新 README.md 与 README_ZH.md，优化 Quick Start 结构，新增 nanobot 集成（飞书聊天机器人） 专章，并提供从配置到部署的一站式教程。
此版本重点提升了“在聊天软件中用 DeepCode 写代码”的体验，方便团队在日常协作中直接使用 AI 编程能力。

This release focuses on better chat-based usage and documentation:
DeepCode + nanobot integration: You can now use DeepCode directly from a Feishu bot, including paper-to-code, chat-to-code, and task status queries, with new real-world Feishu screenshots in the docs.
Docs overhaul: Both README.md and README_ZH.md have been updated with a cleaner Quick Start, and a dedicated nanobot Integration (Feishu Chatbot) section that walks you through configuration and deployment end-to-end.
This version makes it much easier to use DeepCode as a coding assistant directly inside your team’s chat workflow.

DeepCode v1.1.0 重磅更新！新增 Docker 容器化支持，一键部署，开箱即用。让代码分析更简单，部署更便捷！

DeepCode v1.1.0 Major Update! Now with Docker containerization support - one-click deployment, ready to use out of the box. Making code analysis simpler and deployment easier!

DeepCode v1.09 - New UI Update
The new React-based web interface is now available with these improvements:
Interactive workflow - AI can ask follow-up questions during code generation
Streamlined interface - Questions appear directly in the chat for better visibility
Cross-platform - Works on Windows, macOS, and Linux with deepcode command
Better stability - Improved process handling and automatic port management
Run deepcode to try the new interface, or use deepcode --classic for the previous version.

DeepCode v1.09 - 新 UI 更新
基于 React 的新版 Web 界面现已可用，主要改进包括：
交互式工作流 - AI 可以在代码生成过程中提出后续问题
简化的界面 - 问题直接显示在对话中，更易于查看
跨平台支持 - 使用 deepcode 命令在 Windows、macOS 和 Linux 上运行
更好的稳定性 - 改进的进程处理和自动端口管理
运行 deepcode 体验新界面，或使用 deepcode --classic 使用旧版本。

Support for Google Gemini official API:
DeepCode now supports configuration of Google’s official Gemini API, including full support for Gemini‑3‑pro.

支持 Google Gemini 官方 API
已支持配置 Google Gemini 大模型的官方 API，并完成对 Gemini‑3‑pro 的适配与使用。

New Version:

1. We have updated DeepCode’s overall workflow and fixed several issues and bugs previously reported by users.

2. We have improved the robustness of the system and introduced targeted optimizations for models with relatively weaker coding capabilities.

3. Our currently recommended most stable and high‑quality large language models include: Claude 3.5 and above, GPT‑4o and above, and Gemini‑pro.

4. We will continue to iterate, further refining the workflow and robustness, and adapting the system for stable operation across multiple model families.
   Future roadmap:

5. DeepCode will release related technical reports.

6. We will launch automated code verification and execution features.

7. We will keep iterating and advancing the system through continuous technological innovation and adoption.

新版本：

1. 我们已对 DeepCode 的整体 workflow 进行了更新，修复了此前大家反馈的一些问题和 Bug。
2. 提升了模型的鲁棒性，并针对部分代码能力相对较弱的模型进行了专项优化。
3. 当前推荐使用的最稳定且高质量的大语言模型包括：Claude 3.5 及以上版本、GPT‑4o 及以上版本，以及 Gemini‑pro。
4. 我们将持续迭代，进一步优化 workflow 与鲁棒性，并针对多种模型进行稳定性适配。

未来规划：

1. DeepCode 将发布相关技术报告；
2. 上线代码自动化验证与执行功能；
3. 持续迭代升级，引入和落地更多前沿技术。

Full Changelog: v1.0.6...v1.0.7

Full Changelog: v1.0.5...v1.0.6