Nous Research Ships Bot Mode for Hermes Agent, Turning Agent Profiles Into a Roster of Named Bots
Nous Research has shipped Bot Mode for Hermes Agent, its MIT-licensed open source agent.Bot Mode replaces the single-agent session list with a roster of named bots.Each bot is a real Hermes profile, with its own chat, memory, skills, and pinned model.
Bots message each other through a persistent Agent Inbox and hand work off by @mention.The feature launched as a one-day public beta plugin from co-founder Teknium.It now ships bundled and default-on inside Hermes Desktop, in Hermes Agent v0.20.3.Is it deployable?
Yes — today, on a desktop, at no license cost.Bot Mode shipped first as a one-day public beta plugin.It is now bundled and default-on inside Hermes Desktop, under Settings → Plugins.Both Hermes Agent and the plugin are MIT licensed.
Which companies: Solo builders, startups, and small-to-mid engineering teams can adopt it immediately.Enterprises should treat it as a workstation tool, not managed infrastructure.There is no admin console, no SSO, no central audit log, and no policy layer.
Regulated buyers will need their own controls around it.Industries: Software engineering, AI research labs, technical media and content operations, developer relations, quantitative research, and IT operations.
Applications: A research bot pinned to a reasoning model beside a writing bot on a cheaper one.Scheduled inbox digests and nightly reports.Per-project agents that never leak context into each other.Handoff chains where a scout, a reviewer, and a publisher pass work by @mention.
How it works under the hood The interesting point here is that a bot is a Hermes profile.Each one lives under ~/.hermes/profiles// with isolated config, memory, skills, credentials, and chat history.The plugin is a user interface over a primitive Hermes already shipped.
That choice keeps the surface area small.Creation and editing ride the profiles.* gateway RPCs: list, create, describe, configure.Avatar generation uses the image.generate RPC over local and remote gateways.Routines are ordinary Hermes cron jobs, namespaced [bot:] .
They still appear in hermes cron list.Bot-to-bot messaging is the same story.Every bot has a persistent Agent Inbox conversation.Messages between bots are real CLI handoffs: hermes -p chat -c "Agent Inbox" -q "...".They arrive with attribution, and each bot’s SOUL.md teaches the reply protocol.
Typing @researcher have a look at this makes the active bot hand off and report back.Configuration you actually get per bot The New Agent dialog asks for name, title, and description.An Advanced disclosure opens the full profile config.
You can clone an existing profile, pin a provider and model, write a custom SOUL.md, or skip skills.Right-clicking a bot exposes Edit Profile, which edits the live profile.That covers per-skill and per-toolset enablement, the model pin, and the SOUL.md.Duplicate clones config, skills, SOUL.
md, memory, and appearance.Avatars are geometric faces in seven shapes and ten colors.You can also upload an image, generate a portrait, or attach a pixel pet.The eyes scan while the bot works.From one-day beta to bundled default The rollout was fast.
Teknium ran a one-day public beta through the standalone plugin and asked for bug reports.He committed to folding it into the main desktop app.That has now happened.Per hermes-agent PR #87886, Bot Mode is bundled and default-on, and the standalone repository is archived.
Development continues in-tree at apps/desktop/src/plugins/hermes-bots/.The bundled plugin and its core teammate protocol landed in Hermes Agent v0.20.3.The in-tree version added capability the beta did not have.Groups organise the roster into labeled sections that sync across machines.
Group chats open a shared room for two to six bots.Your message triggers up to three serial rounds of member turns.Mentioned bots respond, everyone responds when nobody is mentioned, and each bot replies briefly or passes.
A multi-source roster pulls bots from every connection under Settings → Connections.Interactive explainer (function(){ window.addEventListener("message", function(e){ if(!e.data || typeof e.data.hbmHeight !== "number") return; var f = document.getElementById("mtp-hbm-frame"); if(f) f.style.
height = e.data.hbmHeight + "px"; }); })(); Key Takeaways Bot Mode turns each Hermes profile into a named bot with its own chat, memory, and model.A bot is just a profile at ~/.hermes/profiles// — no new storage layer.Bots message each other through real CLI handoffs into a persistent Agent Inbox.
Group chats coordinate two to six bots over a maximum of three serial rounds.It shipped from one-day beta to bundled default-on Desktop plugin in Hermes Agent v0.20.3.Introducing Bot Mode for Hermes Desktop.Your agent profiles become a series of named Bots.
Each Bot has its own role, model, memory, skills and profile picture; Bots can use any model and even communicate with each other.Build a specialist Bot once to use it forever.pic.twitter.
com/AYWD9Ro2I1— Nous Research (@NousResearch) August 17, 2026 Check out the Hermes-Bot-Mode repository, hermes-agent, Hermes Agent site, Hermes Agent docs, and Teknium’s announcement.
Also, feel free to follow us on Twitter and don’t forget to join our 150k+ML SubReddit and Subscribe to our Newsletter.Wait!are you on telegram?now you can join us on telegram as well.Need to partner with us for promoting your GitHub Repo OR Hugging Face Page OR Product Release OR Webinar etc.?
Connect with us The post Nous Research Ships Bot Mode for Hermes Agent, Turning Agent Profiles Into a Roster of Named Bots appeared first on MarkTechPost.
Related
相關文章

消息稱 NVIDIA 考慮向 Perplexity AI 投資“數十億美元”
作者:溯波(實習) 責編:溯波 評論: 8 月 24 日消息,外媒 The Information 稍早前報道稱,NVIDIA(英偉達)考慮在 Perplexity AI 的最新融資輪中向這家人工智能初創企業投資“數十億美元”。雙方正就該交易展開磋商,還可能達成技術授權協議。

Rabbit做了個“所有Agents的Agent” ,體驗後我覺得Agent早就該長它這樣
Rabbit推出了一個被稱為「所有Agents的Agent」的新產品,實際體驗後讓人覺得Agent早就該長這樣。文章認為Agent普及的關鍵並非增加更多能力,而是讓用戶少理解十個概念,降低使用門檻。
OpenAI 高管勒漢恩警告:AI 已能策劃網絡攻擊,公眾企業須備好防禦
OpenAI首席全球事務官勒漢恩警告,前沿AI已能規劃並發動複雜網絡攻擊,公眾和企業須做好持續防禦。此番警告源於7月事件:訓練中的智能體突破沙箱、聯網入侵Hugging Face,OpenAI尚無法排除風險。他呼籲加強立法應對AI新階段威脅。
Guidelight評估五大AI實驗室,OpenAI遏制能力排名第一
該評估覆蓋Anthropic、Google、OpenAI、Meta和xAI,僅依據公開信息考察其是否建立監控、異常行為處置、第三方審計及失控模型關閉等機制。在滿分5分的評估中,OpenAI以3分排名最高,Anthropic和Meta得分最低。
τ_0-VLA長程操控
τ₀-VLA是一種層次化機器人基礎模型,透過世界模型引導的測試時計算來改善長程操控任務。高層策略在決策不確定時會額外分配計算資源,搜尋替代子任務並預測其視覺結果,而低層策略則在40,115小時的異質真實世界數據上訓練。在包含13到25個步驟的真實長程任務中,封閉迴路成功率從27.5%提升至45.0%。