five changes from the two newest Hermes Agent tags, v0.21.4 and v0.21.5: (1) start a dashboard chat in any project, from your phone (#117834#) the chat tab in the web dashboard has a workspace picker. choose one of your projects or git repos and the new chat starts in that folder. (2) Claude chats can keep their cache for an hour (#118337#) set prompt caching to auto and the chats you type in keep their cache for an hour, so a ten-minute break does not mean writing the whole context again. (3) /branch opens a new thread on Discord, Telegram, Slack and Matrix (#117450#) branch a conversation in a chat app and the copy gets its own thread. the chat you were in stays on the original session. (4) retry when the usage limit resets, in Hermes Desktop (#116288#) when a provider's usage limit stops a turn and says when it lifts, the error card offers to retry at that time. Hermes runs the turn once, with a countdown you can cancel. (5) background commands get a row in the live dock (#117464#) in the CLI and TUI, each command the agent runs in the background shows above the prompt with its latest line of output, next to your subagents.

Hermes Desktop now comes in 9 languages: - English - Simplified Chinese - Traditional Chinese - Japanese - Arabic - Russian - French (new) - German (new) - Spanish (new) Settings > Appearance > Language

people first feedback about this model: fast, to the point, and the reasoning stays short. that matters most in an agent, where every tool call waits on the model. free on Nous Portal right now, with a 1M context and image and video input. `/model stealth/space-bunny-alpha`

your first week with Hermes Agent, one outcome a day, about twenty minutes each: - day 1: install it and have one real conversation - day 2: let it learn how you work - day 3: turn something you repeat into a skill - day 4: reach it from your phone - day 5: give it a job that runs without you - day 6: bring it into a project - day 7: look back, then tidy up bookmark it, read it, point your agent at it

we added 64 new Hermes Agent use cases to the docs page last month, 45 from reddit and 19 from X. it now shows 326, every one linked to the real post. mostly dev workflows, then business ops, creative work and cost-cutting setups. if you're wondering what to build next, start here.

Sometimes all you need is a chat and a sessions bar, not all the developer tools and information For that, Hermes Desktop App now has a Simple Mode! I made a quick video showing how to switch between the two views, and what they look like. Check it out!

last month I asked what would make Hermes simpler for you. one of the clearest asks was a Desktop for people who are not developers. here it is: Simple mode is the chat, your sessions, Capabilities and Messaging, and nothing else on screen. Hermes can still do everything it did, and Advanced brings your full workspace back whenever you want it. thanks for your feedback, this is how we improve our products :)

a reminder: a lot of the frontier models are at half their list price on Nous Portal right now. Grok 4.7, GPT-5.6 Sol and Sol Pro, Gemini 3.8 Flash and 3.7 Flash, Kimi K3. and a good set of models are free: Ling 3.0 Flash, LongCat 2.0, Laguna S and XS, Step 3.7 Flash, Solar Pro 4. /model in any chat shows the current price next to the name, and the ones on sale are marked.

here is what you told me about running Hermes Agent for more than one person: - most of you run one gateway that several people talk to, Telegram at home, Slack at work. "it is like everyone is talking to the same co-worker." - one profile per person on the same box is the other common shape: separate memory, separate config, one machine. - some of you go further and give each person their own container or VM, up to a hundred isolated instances on one company cluster. - you named per-user permissions inside one shared bot as the thing you want most: one bot in a group chat that only uses the tools each person is allowed to use. - several of you built that layer yourselves already, with chat-ID role gates and per-user access checks in front of every turn. - on memory, you agreed on one pattern: each person keeps their own, and the shared part has a single writer. - you also asked for one place to create bots, invite people and assign rights. your answers are clear, and we will use them as we build out Hermes for teams. Hermes Agent community for the win

two Hermes Agent tags went out last week, v0.21.2 and v0.21.3. here are five things in them you can use today: (1) the agent can log you into websites (#106480#, #107585#) say "log into GitHub" and Hermes fills the password from 1Password, Bitwarden or its own encrypted vault. the model never sees it. two-factor codes come from a saved authenticator key, or you type them in when asked and the code never enters the conversation. (2) a check before an expensive model switch (#95267#) switching models in the middle of a long chat means the next reply re-reads the whole conversation at full input price. on a large session Hermes now asks first. (3) cron jobs can run when something happens, not only on a schedule (#95262#) point a webhook route at an existing cron job and the job fires the moment the event arrives, a new PR comment for example. the job keeps its schedule as a fallback. (4) long pastes become a file in Hermes Desktop (#108901#) paste more than 3,000 characters into the composer and it turns into a small attachment chip instead of filling the input box. the text goes along as a file. (5) fix a queued prompt before it runs (#95999#) in the CLI, /queue on its own lists what is waiting for the next turn, and you can edit, remove or reorder the lines before they are sent.

the plugin catalog pages are live. before you install anything, it shows: - the exact commit the maintainers reviewed, the one `hermes plugins install ` checks out - the tools, hooks and environment variables the plugin declares - the README from that reviewed commit, and screenshots if the author added them - the Desktop install button and the CLI command, side by side - a "more by this author" shelf, and every author has a page of their own for plugin authors: the page is built from your catalog entry, so what readers see is the pinned commit. move the pin with a reviewed PR and the page follows.

Hermes Wingtips #79#: `agent[.]disabled_toolsets` Hermes Agent picks its tools per platform: the CLI, Telegram, Discord and the rest each keep their own list of toolsets, set in `hermes tools`. this key is one list above all of them. a toolset named here is removed everywhere at once, on the CLI and on every gateway platform. it is applied after the per-platform lists. `web` takes web_search and web_extract out of every session. `memory` removes the memory tools and also drops the memory guidance from the prompt.

level 5 of my duck's skill tree: one policy that walks and stands up. trained it on my 5090: 20,000 iterations, 4096 simulated ducks, 5 h 26 min. left is iteration 0, right is the trained one. it walks and holds for the full 20 s, and catches a forward lean in half a second.

a habit from this week: before I write a line about a Hermes Agent setting, I ask the install itself. hermes config get approvals.mode it prints the value that install is actually running, default or not, no hunting through config.yaml. the other two I lean on are hermes config set and hermes config unset . I run the set and the unset on a throwaway profile, so trying a setting costs nothing.