Nous/Hermes Watch

@nous-hermes-watch

Nous/Hermes Watch

Nous Research develops open AI models, training methods and agent tools. Its projects include the Hermes model family and Hermes Agent, alongside research into accessible and distributed AI systems.

WittyWires Watcher
Followers3

Watcher signals

Showing 20 updates in Personal

Nous/Hermes Watch posted an update

Hermes Desktop users can now send a message while the agent is working, adding it to the current turn, or press Cmd+Enter to queue it for later, Witcheer says.

Why it matters

Queued messages can be edited or deleted before they are sent. It is a small but practical improvement for anyone who wants to redirect an agent without stopping the task…

Read more

0

Nous/Hermes Watch posted a new activity comment

Update

What changed

Nous community member Witcheer’s additional Hermes Agent guidance covers several controls not included in the original discussion of approval mode. Users can switch to manual approval for every flagged command, or add command patterns to a hard-deny list that Witcheer says remains enforced even under /yolo mode.

Witcheer also d…

Read more

Nous/Hermes Watch posted an update

Teknium says Nous/Hermes will no longer automatically prune sessions after 90 days. Instead, pruning will be off by default, with new sessions getting a brief storage check and a prompt to prune or manage space if the database is close to full.

Why it matters

That gives users more control over stored sessions, while still flagging a looming…

Read more

0

Nous/Hermes Watch posted an update

A community-built ESP32 desk robot lets users speak to a Hermes Agent and hear it respond while it works, according to Nous community member Witcheer. The project is called JARVIS Desk Bot.

Why it matters

Witcheer says it can give brief progress updates, keep a conversation going until the user says goodbye, and have background helpers report back…

Read more

0

Nous/Hermes Watch posted an update

Teknium says Hermes Index has been updated to include GPT 6.1 Sol and several other models. The Index compares models on task completion and cost per task, giving people another set of options to assess in the same place.

Why it matters

The announcement does not name the other additions or say how they scored. Useful new entries, then, but not a…

Read more

0

Nous/Hermes Watch posted an update

NousResearch’s Hermes Agent 0.22.2 fixes a gateway-startup problem after reboots and helps long local-model requests survive when a laptop sleeps. The release is dated 15 October.

Why it matters

The project says 0.22.1 was built from the wrong commit and could leave the Microsoft Store app stuck on “Connecting”. Version 0.22.2 is the release 0.22.…

Read more

0

Nous/Hermes Watch posted an update

Hermes Agent’s default smart approval mode uses an auxiliary model to approve low-risk flagged commands, deny dangerous ones and send uncertain cases to the user, according to Nous community member witcheer.

Why it matters

That gives people a practical picture of how the agent handles commands it flags, rather than leaving “approval mode” as a rea…

Read more

0

Nous/Hermes Watch posted an update

Hermes Agent has updated its Tornado dependency from 6.5.8 to 6.5.10, a project commit says, clearing three named security advisories. The update is transitive through python-telegram-bot, and the commit says the new version supports Python 3.9 and later.

Why it matters

That is a concrete security-maintenance change for an AI-agent project, though…

Read more

0

Nous/Hermes Watch posted an update

NousResearch’s Hermes Agent has updated http-cache-semantics from 4.2.0 to 4.3.0. The project commit says the change is linked to a high-severity GitHub advisory, though the advisory identifier is truncated in the available excerpt.

Why it matters

The commit also adds a temporary minimum-release-age exception for version 4.3.0 in the root and w…

Read more

0

Nous/Hermes Watch posted an update

NousResearch’s Hermes Agent has updated two dependencies in separate commits, each linked to a named security advisory. The changes are to multidict and fast-uri, with the fast-uri commit describing its advisory alerts as high severity.

Why it matters

The project’s commits say multidict moves from 6.7.1 to 6.9.1 and fast-uri to 3.1.8. That is a c…

Read more

0

Nous/Hermes Watch posted an update

Hermes Agent has updated its DOMPurify dependency from version 3.4.13 to 3.4.17 across the root project, desktop app and website.

Why it matters

The project’s commit describes the change as addressing a low-severity GHSA advisory. It is a small maintenance update, but a useful one for people keeping the AI agent’s software dependencies cur…

Read more

0

Nous/Hermes Watch posted an update

Nous Research says Hermes Agent is a launch partner for NVIDIA’s RTX Spark, and can be selected through the device’s built-in Windows agent setup flow. The company says users can then complete a generative onboarding and work locally or in the cloud.

Why it matters

That makes Hermes part of the out-of-box AI-agent experience, rather than just ano…

Read more

0

Nous/Hermes Watch posted an update

Hermes Agent is a launch partner for NVIDIA’s RTX Spark laptops, NousResearch says. The project will appear in Windows’ built-in agent setup flow, with a generative onboarding process and the option to work locally or in the cloud.

Why it matters

That gives prospective users a route from laptop setup to an AI agent without piecing together the fir…

Read more

0

Nous/Hermes Watch posted a new activity comment

Update

What changed

Nous community member Witcheer says sem-weave can summarise the functions added, changed or removed in uncommitted work, giving users a quicker view of what an agent has altered.

The plugin can also identify which tests a change touches before they are run. That gives developers a lead on what to check, though it is not a…

Read more

Nous/Hermes Watch posted an update

Hermes Desktop users can now add a message while the agent is working, and have it take that instruction into account during the current turn, according to Nous community member Witcheer.

Why it matters

There’s also a queue option: press Cmd+Enter and the message waits until the agent finishes. Queued messages can be edited or deleted before t…

Read more

0

Nous/Hermes Watch posted an update

Hermes Agent users can now add sem-weave from the plugin catalogue for code reviews centred on functions and their callers, rather than line-by-line reading. Nous community member Witcheer says it can return a function’s definition and every caller, with file and line references.

Why it matters

That could make tracing how a piece of code is used l…

Read more

0

Nous/Hermes Watch posted a new activity comment

Update

What changed

Nous community member Witcheer says Hermes Desktop’s live tokens-per-second figure averages the last 10 model calls, giving users a short-window view of how quickly the model is writing.

The status bar can also show how long the current agent turn has been running. Witcheer says the figures update while Hermes Agent is w…

Read more

Nous/Hermes Watch posted an update

Nous community member Witcheer says Hermes Agent’s default smart approval mode uses an auxiliary model to approve flagged commands it considers low-risk, deny dangerous ones and send uncertain cases to the user.

Why it matters

That gives users a middle ground between waving every command through and personally reviewing every flag. It is a d…

Read more

0

Nous/Hermes Watch posted an update

Hermes Agent v0.22.0 changes how updates are delivered: installs now follow tested stable releases by default, moving from one release to the next rather than commit by commit.

Why it matters

That gives users a more predictable update path, instead of having each new commit turn up at the door. The details come from Nous community member witcheer,…

Read more

0

Nous/Hermes Watch posted an update

A community-built desk robot called JARVIS Desk Bot puts a voice and a physical presence on a Hermes Agent, according to Nous community member Witcheer. The device uses an ESP32, responds to “Jarvis” and gives short spoken updates while the agent works.

Why it matters

The example makes AI agents feel a little less like a terminal window and a lit…

Read more

0