Hermes3D, Kanban & 'Beats Grok Bot': Fact-Checked

Five distinct claims ran through this week's batch of Hermes Agent coverage. We checked each one against a primary source, Nous Research's own GitHub organisation, release tags and documentation, plus xAI's own Grok Bot announcement. Two of the five claims are genuinely real but mischaracterised (Kanban is real but three weeks old, not new; Bot Mode itself is real and current). Three are community projects being presented without that context (Hermes3D, the email agent, Gemini Notebook integration). One comparison ("beats Grok Bot") has no official source at all.

• Hermes3D: a 212-star independent repository, not a Nous Research feature, whose own README disclaims official affiliation.
• Kanban agent: official and real, but shipped 3 August 2026, not this week.
• "Beats Grok Bot": an unsourced creator claim; no official benchmark or comparison exists from Nous Research or xAI.
• Email agent, Gemini Notebook: community-built skills layered on Hermes Agent's existing capabilities, not Nous Research products.
• What's actually current: Hermes Agent v0.20.5 (19-21 August), the same release covered in our prior story, remains the latest official tag as of 25 August.

Between 24 and 25 August 2026, eight separate videos from the same creator covered Hermes Agent, more than any other single topic in that window across all seventeen channels this site tracks. The titles alone give a sense of the framing: "This NEW Hermes Update Makes Ideas AUTOMATIC", "How to Build an AI Email Agent With Hermes", "New Hermes Agent Just Changed EVERY THING!", "Hermes3D Is INSANE: AI Agents Now Have a 3D Office", "NEW Hermes Agent OS JUST Changed AI AGENTS Forever!", "NEW Hermes Agent Beats Grok Bot!?", "Hermes + Gemini Notebook is ABSURD", and "New Hermes Kanban Agent Is ABSURD!"

That volume and cadence, several videos a day covering incremental angles on the same base tool, is itself worth naming as a pattern. It is a familiar creator strategy (fresh video titles drive more views than one comprehensive explainer) and it is not inherently dishonest. But it does mean each individual video has an incentive to frame its specific angle as the newest, biggest thing, which is exactly where the gap between "genuinely new this week" and "new to this video series" tends to open up. Our prior review covered an almost identical pattern just one day earlier, with a different set of claims (subagents, "Agent OS" branding, free access) drawn from the same creator's content.

"Hermes3D Is INSANE: AI Agents Now Have a 3D Office" describes a live, open-source, MIT-licensed 3D visualisation where multiple Hermes agents appear as characters in a shared office, handing off tasks and working in view. Every factual detail in that description checks out against the source: the project genuinely exists, is genuinely MIT-licensed, and does genuinely connect to Hermes Agent as its backend. What the video does not make clear is who built it and how official it is.

The repository lives at github.com/iamlukethedev/Hermes3D, maintained by an independent developer going by "Luke The Dev," who spun it off from an earlier, unrelated project called Claw3D built for the separate OpenClaw agent. As of this writing it has 212 stars, 4 contributors and 1 open issue, a modest hobby-scale project by any measure, especially set against the parent Hermes Agent repository's 236,000 stars and roughly 3,000 contributors. Most tellingly, the Hermes3D repository's own README states plainly that it is "not affiliated with, endorsed by, or maintained by the teams behind the agent backends it connects to," a direct, self-declared disclaimer that this is a third-party integration, not an official feature.

None of this makes Hermes3D a bad project, small, independent, community-built tools that visualise or extend a larger open-source agent are a genuinely healthy sign of an ecosystem, and the same is true of tools built around Claude Code, Codex and other major agent harnesses this site covers. The issue is purely one of framing: presenting a hobbyist's side project as "AI agents now have a 3D office," without noting that Nous Research had no hand in building it, leaves a viewer with a materially wrong picture of what Hermes Agent itself, as shipped by its actual maintainers, currently includes.

"New Hermes Kanban Agent Is ABSURD!" describes a kanban-style board where specialised agents plan, build and review an entire project from a single sentence. Here the underlying feature is genuinely official, Nous Research really does ship a Kanban board for Hermes Agent, but the "new" framing does not hold up against the release history.

Kanban shipped as the founding plugin of Hermes Agent's new plugin SDK, part of v0.20.0, "The Herald Release" (GitHub tag v2026.8.3), published 3 August 2026, over three weeks before this week's video covered it as a fresh capability. It is documented in detail at Nous Research's own docs site, under hermes-agent.nousresearch.com/docs/user-guide/features/kanban, alongside the rest of Herald's plugin architecture. Herald itself was, by Nous Research's own description, the largest release in the project's history at the time, adding real-time streaming voice with barge-in, Agent-to-Agent protocol v1.0, and the plugin SDK Kanban launched alongside, details covered in full in our Herald release review.

A three-week-old feature getting covered as if brand-new is a milder version of the Hermes3D problem: not a fabrication, but a framing gap that overstates how current the content actually is. For anyone tracking Hermes Agent's real release cadence, Kanban belongs in the Herald story from early August, not in a "this week's big Hermes news" roundup published three weeks later.

"NEW Hermes Agent Beats Grok Bot!?" is the most direct comparative claim in this week's batch, and it is worth separating into its two component parts: is Grok Bot real, and did anyone official actually say Hermes Agent beats it?

Grok Bot is real. xAI announced it on 11 August 2026 as an "always-on AI teammate", agents that get their own persistent cloud computer rather than running only inside a chat session, initially in beta for SuperGrok Heavy and Cursor Ultra subscribers, with access expanded to additional plans on 21 August 2026. That much is a genuine, dateable xAI product launch, our own Grok Bot explainer covers what it actually does.

The "beats" part is where the claim breaks down. Neither Nous Research nor xAI has published any head-to-head benchmark, comparison or statement addressing the other's product. The framing traces specifically to independent creator content, comparison posts and videos published by the same creator behind this week's cluster, not to a documented evaluation either company has put its name to. That does not mean a fair, independently-run comparison between Bot Mode and Grok Bot would necessarily favour one or the other; it means no such comparison currently exists in a form that can be verified, and "beats" in a video title should be read as marketing framing rather than a tested result until one does.

Two further videos, "How to Build an AI Email Agent With Hermes" and "Hermes + Gemini Notebook is ABSURD", describe genuinely useful workflows: an agent that finds leads, verifies addresses and drafts outreach campaigns from a dashboard, and a setup that pairs Hermes with Google's NotebookLM (referred to in the video as "Gemini Notebook") for research and content work. Neither exists as an official repository under the NousResearch GitHub organisation.

Both are best understood as skills, in Hermes Agent's own terminology for the extensible, community-contributable capabilities its skill system supports, built by the wider creator community and popularised through tutorial content rather than shipped by Nous Research itself. This is, again, not a criticism of the workflows themselves; building an email-outreach skill or a NotebookLM bridge on top of a general-purpose agent harness is a legitimate and often genuinely useful thing to do, the same pattern this site has flagged with other creator-built "Agent OS" setups. It just means these are demonstrations of what a user can build with Hermes Agent's existing, general-purpose tool-use and skill system, not new capabilities Nous Research itself added this week. Worth flagging directly: Nous Research's own NVIDIA SkillEvaluator Tier 1 advisory scanning, added in v0.20.4, exists specifically to scan third-party skills like these for known issues at install time, precisely because the skill ecosystem is open and community-built rather than centrally vetted.

Hermes Agent is a genuinely popular, actively developed, MIT-licensed open-source project with 236,000 GitHub stars and a real, fast release cadence, all reasons it attracts this much creator attention in the first place. That popularity is also exactly what makes it fertile ground for exactly this kind of framing drift: a large, active ecosystem produces a steady stream of genuinely interesting community tools, skills and integrations, and covering all of them as if they were the maintainer's own roadmap is an easy, low-effort way to generate a constant supply of "NEW" video titles without waiting for Nous Research's actual, comparatively slower release schedule.

The practical risk for readers is less about any single video and more cumulative: someone who only follows this kind of coverage over several weeks ends up with a mental model of Hermes Agent that is a blend of real releases, three-week-old features re-presented as current, and other developers' independent side projects, all flattened into one undifferentiated stream of "Hermes just shipped X." None of the individual pieces are necessarily false, but the aggregate picture misleads by omission.

Checked directly against Nous Research's own GitHub releases page, the tag history since our last review has not moved: v0.20.5 (tag v2026.8.19, 19-21 August 2026) remains the latest official release as of 25 August. There is no v0.20.6 and no v0.21.0. That release, covered in full in our prior piece, includes Bot Mode's group-room threads and foldable summaries, a keyless five-provider web-search tier, drag-and-drop file attachments, CLI improvements, and the execution-discipline and stall-guard reliability fixes credited to Composio's eval findings.

In other words, the single most current, verifiably official thing about Hermes Agent this week is the same release this site already covered on 24 August. Everything genuinely new in this week's video batch (Hermes3D, the email agent, Gemini Notebook integration) is community-built rather than Nous Research-shipped, and the one item from Nous Research's own roadmap being presented as new (Kanban) actually shipped three weeks earlier.

• Star counts and contributor figures change over time. Hermes3D's 212 stars and Hermes Agent's 236,000 are snapshots as of 25 August 2026, not fixed numbers; a community project can grow quickly.
• Absence of an official comparison is not the same as a false claim. We could not verify "Bot Mode beats Grok Bot," but we also could not verify the reverse; the honest position is that no independently checkable comparison currently exists.
• Community skills are genuinely usable. The email agent and Gemini Notebook workflows are real, working setups a reader could build; our finding is about their provenance (community, not Nous Research), not their usefulness.
• This site cannot audit Hermes3D's code for security or quality. A 4-contributor, 212-star repository has had far less scrutiny than Hermes Agent's own 3,000-contributor codebase; treat third-party skills and integrations with the same caution this site has flagged for Hermes' skill ecosystem generally.

The check used throughout this piece is simple and repeatable: for any "NEW" claim about an open-source agent tool, go to the maintainer's own GitHub releases page and look for a tag with a matching or preceding date. If the feature shown in a video does not appear in the official changelog at all, search GitHub directly for the tool's name, if what turns up is a different, smaller repository than the main project, that is a strong signal the feature is a community add-on rather than an official one. Star count, contributor count and issue count are quick, reliable signals for distinguishing a maintainer's flagship repository from a hobbyist's side project pointed at the same underlying tool.

For Hermes Agent specifically, the two sources worth bookmarking are github.com/NousResearch/hermes-agent/releases for the actual, dated release history, and hermes-agent.nousresearch.com/docs for the maintained feature documentation. Anything claimed as new that doesn't appear in either is either very recent (worth checking back on) or not an official Nous Research feature at all.

Hermes Agent remains a genuinely capable, fast-moving, MIT-licensed open-source project, and nothing in this piece changes that assessment. What this week's video cluster gets wrong is not the existence of Hermes3D, the Kanban board, or the email and Gemini Notebook skills, all of those are real and, in Kanban's case, officially shipped. What it gets wrong is presenting a hobbyist's 212-star side project, a three-week-old feature, and two community skills as a single wave of brand-new Nous Research capability, with an unsourced "beats Grok Bot" claim layered on top for good measure.

The most current, genuinely official thing Nous Research has shipped remains v0.20.5 from 19-21 August, exactly what our prior review already covered. Everything else in this week's batch is worth knowing about on its own terms, as community tools and creator workflows built on a capable open platform, just not as this week's product roadmap from the people who actually build Hermes Agent.