agents.changelog

official sources · ongoing coverage

Release history

All retained Hermes Agent updates

Hermes Agent update feed

Hermes Agent v0.21.6

About this release: Also in the window, undocumented here on purpose: live transcription while you speak on CLI, TUI and Desktop (stt.streaming); spoken voice turns routed to their own model (auxiliary.voice_chat) with reasoning off by default; third-party plugins running in a per-profile plugin host (plugins.isolation: host) and one Settings ▸ Plugins home for their settings pages; favorite and pinned models at the top of the Desktop picker, a usage chip before a subscription hits its wall, and rate-limited providers that say why and until when; local models in hermes model plus llama.cpp CUDA on Linux x64 and arm64; pluggable, layered language packs across core, Desktop and TUI; a summary-first hermes status (--short, --full); a review-pane diff scope selector (Uncommitted / Branch / Last turn) and keep-awake that follows the turn; cron, doctor and home-channel warnings when the cron store goes unwritable; approval flags for credential uploads and invisible Unicode in terminal commands; a Discord setup that checks the bot token and prints the invite link; a unified Kanban workflow definition; activate.fish; GPT-6.1 Sol and Claude Sonnet 5.5 / Haiku 5.5 in the catalogs; and dozens of new community plugins.

解读这一条是本次发布的功能大盘点,塞进了大量面向用户的新能力,可以按使用场景理解: - 语音相关:在 CLI、TUI 和桌面端可以“边说边转写”的实时语音输入(stt.streaming);语音对话会走独立模型(auxiliary.voice_chat),默认关闭推理以获得更快响应。 - 插件体系:第三方插件可以跑在按配置隔离的插件宿主中(plugins.isolation: host),降低插件互相干扰或影响主程序的风险;桌面端新增统一的“设置 ▸ 插件”页面来管理它们的配置。 - 桌面端体验:模型选择器支持“收藏/置顶”模型,常用模型排在最上面;订阅额度快用完前会显示用量提示;被限流的供应商会明确告诉你原因和何时恢复。 - 本地模型:hermes model 可直接管理本地模型,Linux x64 和 arm64 上支持 llama.cpp 的 CUDA 加速。 - 可定制与可观测性:核心、桌面端、TUI 支持可插拔的分层语言包;hermes status 增加摘要优先输出(--short / --full),快速看状态不用翻一大屏。 - 代码审查与终端:审查面板新增 diff 范围选择器(未提交 / 分支 / 最近一轮对话),方便只看你关心的改动;“保持唤醒”会跟随一次对话的起止,不会一直亮着。 - 运维与安全提示:定时任务存储变为不可写时,cron、doctor 和频道会给出告警;上传凭证、以及终端命令中包含不可见 Unicode 字符时都需要额外审批。 - 集成与生态:Discord 初始化会校验机器人 token 并直接打印邀请链接;统一的 Kanban 工作流定义;新增 activate.fish 方便 fish shell 用户;模型目录加入 GPT-6.1 Sol、Claude Sonnet 5.5 / Haiku 5.5,以及数十个新的社区插件。 简单说:这次更新让语音交互、插件隔离、模型管理与审批安全等日常使用环节都更顺手,具体行为都以“功能默认可用、风险操作需确认”的方式落地。

Hermes Agent v0.21.5 (v2026.9.24)

no notable user-facing improvements in this release

解读此版本无值得关注的用户可见改进

Hermes Agent v0.21.4 (v2026.9.21)

About this release: Also in the window, undocumented here on purpose: a host-wide gateway singleton lock with a rendezvous record, and Desktop attaching to the running host backend instead of spawning a second one; one backend-owned connector operation with a setup card on Desktop, TUI and CLI; --format stream-json structured JSONL output for the CLI; skills.autoload pinning skills into every new session's prompt; a Desktop chat/UI font picker, one-click local engine updates, and plugin uninstall from the Plugins hub; a decline unauthorized-DM behavior for the gateway; a configurable MCP discovery connect cap (mcp.discoveryconcurrency); session_search after/before bounds plus OR-relaxed recall retry; hermes sessions set-journal-mode; LTX 2.5 and Kling O3 in the video catalogs; a website page for every catalog plugin and author, with pinned-commit READMEs and added/updated sorting; a dozen new community plugins in the catalog (tailscale, ssh, shodan, terminal, rss, resetwatch, done-bell, kiwi, cognee, Octen); and a large run of profile/multiplex isolation, cron, kanban, Desktop and state.db fixes.

解读这一版集中加入和增强了一批实用能力:CLI 新增 `--format stream-json`,可输出结构化 JSONL,方便脚本、CI 或自动化工具解析;`skills.autoload` 能把指定技能自动注入每个新会话的提示词;Desktop 增加聊天/UI 字体选择、本地引擎一键更新、插件中心一键卸载插件,并能连接到已运行的后端而不是重复启动;网关新增全机单例锁、拒绝未授权私信等行为;MCP 发现连接数可配置(`mcp.discoveryconcurrency`);`session_search` 支持按时间前后范围检索并带 OR 放宽重试;还加入 `hermes sessions set-journal-mode`、视频目录的 LTX 2.5 / Kling O3、插件网站页面和大量社区插件。同时也有大量隔离、cron、看板、Desktop 与状态数据库修复。对使用者来说,这意味着自动化集成更顺手、多实例/网关管理更可靠、插件生态更丰富,桌面端日常操作也更方便。

Hermes Agent v0.21.3 (v2026.9.14)

About this release: Also in the window, undocumented here on purpose: server→client JSON-RPC requests and a Pydantic wire-contract registry with generated TS/OpenRPC for the TUI/Desktop gateway ; reasoning-effort selection on every model picker, a composer pill and per-auxiliary control in Desktop; OpenRouter OAuth PKCE login; HEIF/HEIC/AVIF image decoding; the Honcho peer-model setup rework; MCP OAuth refresh tokens bound to their issuer; a daily MCP re-auth nudge in Desktop; Wan 3.0, Kling 3.0 / Kling Image v3, MiniMax H3 Max Turbo, Gemini Omni Flash 1.1 and Meta Muse in the FAL catalogs; Slack pasted tables and the Agent Sessions API; multiplexed-profile isolation and gateway-liveness fixes; and the state.db WAL refusal on cross-VM filesystems.

解读这条虽然写着“此处故意不展开”,但实际列出了一批明确的新功能与能力增强,属于功能更新。值得关注的几点:① 网关新增“服务端→客户端”的 JSON-RPC 请求,并引入 Pydantic 定义的通信契约注册表,可自动生成 TypeScript / OpenRPC 类型,让 TUI、Desktop 与后端之间的接口有统一、可校验的定义,减少手写类型不一致导致的 bug;② 每个模型选择器都能选推理强度(reasoning effort),Desktop 里还有 composer 上的快捷药丸和每个辅助模型的独立控制,简单任务可以调低思考量省时省钱,复杂任务再调高换取质量;③ 支持 OpenRouter 的 OAuth PKCE 登录,用浏览器授权代替手动粘贴 API Key,更安全也更好用;④ 支持 HEIF / HEIC / AVIF 图片解码,iPhone 拍的 HEIC 照片可以直接丢进来,不用先转成 PNG/JPEG;⑤ 模型目录新增 Wan 3.0、Kling 3.0 / Kling Image v3、MiniMax H3 Max Turbo、Gemini Omni Flash 1.1、Meta Muse 等;另外还有 Slack 粘贴表格、Agent Sessions API、MCP OAuth 刷新令牌绑定签发方、Desktop 每日 MCP 重新授权提醒等改进。

Hermes Agent v0.21.2 (v2026.9.11)

Password-blind credential vault: The agent can now sign in, pay, and fill addresses from 1Password, Bitwarden, or the local Hermes vault without ever seeing a secret; two-factor codes come from a saved authenticator key or are asked for in the user's UI . Private git plugins install with the user's stored credentials .

解读这次加入了一个“看不见密码”的凭据保险库能力:Agent 可以帮你登录网站、完成支付、自动填写收货地址,但这些凭据都保存在 1Password、Bitwarden 或本地 Hermes 保险库中,模型本身始终接触不到明文密码。双重验证码来源有两个:要么从你预先保存的验证器密钥自动生成,要么在用户界面上向你索取,因此密钥也不会进入对话上下文。另外,私有的 Git 插件仓库也能直接使用你已保存的凭据来安装,不用再手动输入 token。为什么重要:密钥一旦进入模型上下文,就可能被提示注入、日志或截图泄露;现在改为由保险库代为执行操作,只是返回“已完成”的结果。实际使用中,比如你对 Agent 说“帮我登录某某后台下载上月账单”,它会让保险库去填表单,而不会把密码念出来。