v0.6.20

Highlights

• Workflow is now online: create agent nodes, connect them into a graph, and run the full flow directly from the Workflow page.
• Workflow runs now persist records, graph snapshots, per-node sessions, live node status, failure reasons, and run stop/delete actions.
• Workflow viewing and editing are more complete with run history, read-only snapshots, per-node chat session review, larger connection handles, and a create-workflow-first guard.

Fixes and Runtime

• Desktop launches now let agent-browser resolve Chrome from AGENT_BROWSER_HOME instead of pinning a stale executable path, with patch integrity guards and browser stdout fixes.
• Packaged browser runtimes now pin Chrome for Testing 149.0.7827.55 on supported desktop platforms to avoid runtime browser regressions.
• ESP32 and Xiao Fang He details were polished with updated tutorial links, a lighter badge background, and a Windows-compatible flowing badge animation.

Validation

• npm run harness:check
• npm run build

v0.6.19

更新内容

• 本次版本包含 0.6.18 之后合并的 16 个 PR，重点改进前端可见的设备、语音、Coding Agent 和设置体验
• Coding Agent 支持在会话中切换模型，新建会话也会正确跟随全局默认模型
• MCU 交互更稳定：修复开机 Wi-Fi/boot 状态闪动、音频只发送到目标设备，并优化实体按键双击清空与打断体验
• TTS 配置改为按 profile 保存，音速、音调和测试语音会读取同一套配置来源
• Coding Agent 的模型切换不会影响既有会话，并修复自定义 provider key 透传
• 补齐 skills runtime 过滤、定时任务排序、技能详情高度限制、Windows 更新恢复、配置备份兜底和 logo 资源压缩
• Web UI Release 包现在会包含已同步的 ESP32-C3 固件产物

v0.6.18

更新内容

• Hermes Studio MCP 拆分为 api、devices 和 use 工具集，新增 session、model、chat-run 与 ops 相关操作
• 聊天和群聊历史更可靠：UI 历史与模型上下文分离，空消息处理更安全，run failed 不再覆盖已有助手回复，reasoning 回放也更一致
• MCU 设备反馈更清楚：工具调用会优先显示在语音之上，音频不会过早标记任务完成，活跃设备登录更稳定，ONLINE/IP 长文本支持滚动
• 语音和消息工作流新增自定义 HTTP TTS/STT 端点、音频转码、fork 消息操作、持久化 fork 标记，以及更清晰的上传错误
• 桌面与 runtime 设置更安全：Windows PATH 注册支持 Unicode，支持发布 Hermes Agent 0.17.0 runtime，并补齐移动端 PWA 元数据和多项 profile/auth 恢复修复
• 各项细节的优化

hermes-0.17.0-runtime

Fix Windows PATH shim Unicode handling (#1695)

v0.6.17

Changelog

• 修复 Codex 和 Claude Code 的对话启动问题。

v0.6.16

Highlights

• Bumped Web UI and desktop package versions to 0.6.16.
• Added the 0.6.16 frontend changelog entry and localized changelog strings across all client locales.
• Updated the website Web UI version manifest with 0.6.16.
• Provider settings now preserve custom API modes, cover LongCat correctly, support deleting v12 provider entries, and keep model catalogs in sync with auth and API-key image providers.
• Chat commands now include reload-skills, web title updates, unknown slash passthrough, and a clean session context API without tool calls or tool results.
• Session navigation, file drag-and-drop uploads, symlinked skill discovery, packaged static asset delivery, Linux desktop viewport sizing, and live compression status rows were polished.
• Added outbound relay and MCU login flow support, plus a Web UI CLI --no-open flag.

Validation

• Release created from release/frontend-0.6.16-changelog at e8b994b.
• npm run openapi:generate
• npm run test -- tests/client/i18n-coverage.test.ts
• npm run harness:check
• git diff --check

v0.6.15

Highlights

• Chat sidebars were redesigned, with clearer history loading controls for long conversations.
• Added the skill command picker for discovering and inserting available skill commands from chat input.
• Added Claude and Gemini OAuth providers, normalized Gemini model IDs, and added model selector refresh for config.yaml changes.
• Refreshed Thinking indicators and toolbar layout so reasoning progress is easier to read without interrupting message flow.
• Restored visible dark-theme borders for input and select form controls.
• Improved runtime startup checks, desktop self-update handling, gateway shutdown/respawn behavior, and desktop server shutdown cleanup.
• Updated README and website showcase media.
• Removed inline clear controls from the auxiliary model editor fields.

Validation

• Release created from main at f3365ce.
• npm run build passed for the auxiliary model editor change before merge.

v0.6.14

What's Changed

• Redesign website landing page by @EKKOLearnAI in #1471
• Force website light theme by @EKKOLearnAI in #1473
• [codex] Add Hermes Studio client manual docs page by @thursdaycapital in #1475
• feat: add completion notifications by @EKKOLearnAI in #1477
• fix(profiles): reject reserved rename targets by @qdivan in #1478
• feat: add custom desktop title bar by @EKKOLearnAI in #1480
• feat: add workspace folder actions by @EKKOLearnAI in #1484
• fix: stabilize Linux deb desktop launch path by @EKKOLearnAI in #1485
• Remove legacy gateway manager cleanup by @EKKOLearnAI in #1486
• Avoid Hermes bridge checks for coding agent resumes by @EKKOLearnAI in #1487
• [codex] fix group chat agent and member identity by @EKKOLearnAI in #1491
• [codex] fix static asset compression by @EKKOLearnAI in #1492
• [codex] secure device pairing requests by @EKKOLearnAI in #1493
• [codex] fix desktop Claude Code detection by @sir1st in #1495
• [codex] Improve write gate and desktop startup UX by @EKKOLearnAI in #1496
• [codex] update runtime version management by @EKKOLearnAI in #1499
• [codex] fix write approval settings runtime support by @EKKOLearnAI in #1501

Full Changelog: v0.6.13...v0.6.14

v0.6.13

Changes

• Alibaba Coding Plan now includes the qwen3.7-plus model preset.
• File handling is safer with literal-dot path validation and malformed skill-import filename rejection.
• Chat is more stable on mobile and in browsers without speech synthesis support.
• Agent failures now preserve the bridge error tail, and Claude Code root-permission launches are more reliable.
• Desktop updates are more reliable: Windows upgrades close stale Hermes Studio processes, Cloudflare is checked first, and GitHub remains the fallback feed.
• Release automation now keeps Web UI and Docker releases out of GitHub Latest while full desktop releases are manually promoted to Latest.

v0.6.12

Highlights

• Coding Agent is now a fuller chat experience with dedicated sessions, clearer history, preserved tool/final output, and better chat-state recovery.
• Claude Code and Codex launch more reliably from the desktop app on Windows, with corrected command resolution and readable failure messages.
• Coding Agent reasoning, queue status, and error states are easier to follow with stored reasoning content and a nonblocking queue overlay.
• Desktop runtime version management supports Coding Agent workflows with active-version switching, website version manifests, release packaging support, and desktop-only visibility.

Frontend Updates

• Per-session reasoning effort controls are available for compatible chat models and hidden for Coding Agents where they do not apply.
• Group chat mention positioning, active Profile uploads, file downloads, Markdown previews, and highlighted code block wrapping were improved.
• The changelog modal is taller and the latest changelog entries are localized across supported languages, with 0.6.11 wording cleanup in non-English locales.

Validation

• npm run test -- tests/client/i18n-coverage.test.ts
• npm run build