AgentsAutonomy & tool use
Grok Bot, ChatGPT Dots and Hermes Agent compared: two of the three always-on agents are weeks old and none has been independently evaluated
A creator's hands-on comparison favors xAI's Grok Bot for non-technical users and Nous Research's open-source Hermes Agent for technical ones, and advises waiting on ChatGPT Dots. Both commercial agents are weeks old, and no independent evaluations exist.
AT Pulse research deskOctober 6, 2026 · 5 min read · AI-assisted analysisThe two commercial agents are very new. xAI's Grok Bot entered beta in August 2026, and OpenAI's Dots began rolling out at DevDay on 29 September. Grok 4.7's benchmark gains are vendor-reported, and xAI used different reasoning settings across evaluations. The figure of 418,000 weekly Grok Bot users came from an xAI presentation reported by Bloomberg. [3] [4] [5] [6] [8]
The video's verdict is a matter of fit, not a single winner. It calls Grok Bot the easiest to use, Hermes the most flexible but prone to bugs you have to fix yourself, and Dots promising but unreliable. The Dots problems it describes (slow replies, unanswered calls, and a freeze while operating its own screen) are one user's experience and have not been confirmed elsewhere. [2] [12]
The agents are built and controlled in different ways. Grok Bot's bots all share one cloud machine, along with its files and logins, and pause for approval before sending, publishing, buying or deleting. Dots lets users set rules for which actions need approval, but a dot's memories can only be removed by deleting the whole dot. Hermes is MIT-licensed, works with any model provider, and supports Docker, SSH and Modal sandboxes. [3] [4] [9] [10]
Cost and integrations differ sharply. Grok Bot costs $120–$300 per seat per month and has 220 plugins. One Dot is bundled with ChatGPT Pro and Business Premium, connects to more than 4,000 apps, and its pricing after the first month is unpublished. Hermes has no licence fee, but users pay for the models and hosting they choose. [4] [8] [9] [10]
launch coverage confirms pricing, plugin counts and architecture, but performance and usage figures trace back to the vendors.
Read the full assessment
Implication: teams handing these agents their logins should pilot carefully and check approval gates and credential isolation before relying on a creator's rankings.
Executive brief
The key fact for buyers is that two of these three "always-on" agents are only weeks old. xAI's Grok Bot launched in beta on 11 August 2026, and OpenAI's ChatGPT Dots began rolling out on 29 September. Neither has independent benchmarks. The Reddit post and its video (6 October) are a creator's hands-on opinion, and both promote paid communities. The creator's advice is to pick Grok Bot if you're non-technical and Nous Research's open-source Hermes Agent if you're technical, and to hold off on Dots for now. The Dots reliability complaints, including 30-second delays and unanswered calls, come from one user and have not been independently confirmed.
What changed and event timeline
Grok Bot enters beta
Agents share one cloud Linux machine per account and can delegate to each other. Access is limited to SuperGrok Heavy, Cursor Ultra and Cursor Teams Premium subscribers (;).
Early usage figure reported
Bloomberg reported 418,000 weekly users, up 24% week over week. The number came from a presentation at a London event and was reported on 22 September ().
Grok 4.7 released
xAI says it trained the model to work natively with the Grok Bot harness. All benchmark gains cited are vendor-reported (;).
OpenAI launches Dots at DevDay
The agents run on GPT-6 Astra, each with its own cloud computer, and roll out to Pro and Business Premium subscribers ().
Creator comparison published
The video ranks Grok Bot easiest, Hermes most flexible and Dots "not there yet" ().
Capabilities and access
- Costs $200/month through Cursor Ultra, $120/seat through Cursor Teams Premium, or comes with SuperGrok Heavy at $300/month.
- ChatGPT Dots: Runs on GPT-6 Astra. Connects to more than 4,000 apps and works in Slack and Teams (SiliconANGLE).
- Hermes Agent (Nous Research): MIT-licensed and works with any model provider (GitHub).
Read the full section
- Grok Bot (xAI): In beta. Costs $200/month through Cursor Ultra, $120/seat through Cursor Teams Premium, or comes with SuperGrok Heavy at $300/month. Has 220 plugins and pauses for approval before sending, publishing, buying or deleting (VentureBeat; Reworked).
- ChatGPT Dots: Runs on GPT-6 Astra. One agent is included with Pro or Business Premium; Pro users in the EEA, Switzerland and UK are excluded at launch. Connects to more than 4,000 apps and works in Slack and Teams (SiliconANGLE).
- Hermes Agent (Nous Research): MIT-licensed and works with any model provider (GitHub).
Technical analysis for researchers and developers
- Grok Bot architecture: A "Chief of Staff" bot hands work to specialist bots in shared threads.
- Dots controls: "Proactive research" runs read-only, and Custom Rules decide which actions need approval (DataCamp).
- Hermes: Includes memory that the agent curates itself, automatic skill creation, parallel subagents, cron scheduling, and sandboxes including Docker, SSH and Modal (GitHub).
Read the full section
- Grok Bot architecture: A "Chief of Staff" bot hands work to specialist bots in shared threads. All bots on an account share one VM, files and logins, so they are not separate security sandboxes (Reworked). At launch, backend model routing was automatic and couldn't be chosen by the user (VentureBeat).
- Dots controls: "Proactive research" runs read-only, and Custom Rules decide which actions need approval (DataCamp).
- Hermes: Includes memory that the agent curates itself, automatic skill creation, parallel subagents, cron scheduling, and sandboxes including Docker, SSH and Modal (GitHub).
- Evaluation: None of the three has a reproducible, independent agent evaluation.
Claims and evidence
- Grok 4.7 improved on Terminal-Bench 4.0 from 20.3% to 38.0%
- 418,000 weekly Grok Bot users
- Dots is slow, sometimes doesn't answer, and froze while using its own screen
Read the full section
| Claim | Status |
| Grok 4.7 improved on Terminal-Bench 4.0 from 20.3% to 38.0% | Vendor-reported; xAI used different reasoning settings across evaluations (Data Studios) |
| 418,000 weekly Grok Bot users | Bloomberg, based on an xAI presentation (PYMNTS) |
| Dots is slow, sometimes doesn't answer, and froze while using its own screen | One creator's anecdote (video, 06:56); no independent corroboration found |
| Hermes is the only one of the three that lets you swap the model per worker | Partly supported: Hermes supports switching models (GitHub), and Grok Bot routing was not user-selectable at launch |
Context and prior work
The "chief of staff" orchestration pattern was already being discussed alongside Grok 4.6 in August (blog analysis). Meta's consumer agent Muse launched on 8 September (SiliconANGLE). Bloomberg Intelligence estimated Muse was downloaded about four times as often as Grok Bot (PYMNTS). The video also says Claude already covers multi-agent teams and background routines, though not bundled as one always-on agent (07:23).
Limitations, safety and contested findings
- Which model Grok Bot uses is disputed.
- Grok Bot's shared VM is described as having "a real blast radius," and it lacks native HR and ITSM plugins (Reworked).
- Dots: OpenAI says the agents "can still make mistakes."
Read the full section
- Which model Grok Bot uses is disputed. The video says Grok Bot runs on Grok 4.7 (00:47), while launch coverage described automatic routing with no model disclosed.
- Grok Bot's shared VM is described as having "a real blast radius," and it lacks native HR and ITSM plugins (Reworked).
- Dots: OpenAI says the agents "can still make mistakes." You can't delete a dot's memories without deleting the whole dot, and pricing after the first month hasn't been published (DataCamp).
- Hermes needs troubleshooting by the user, for example with
hermes doctor(DataCamp). The video makes the same point about patching bugs yourself (05:37). - The source is commercial. The video repeatedly promotes the creator's paid communities (03:15).
Business and practitioner implications
- Cost: Grok Bot costs $120–$300 per seat per month. Dots is bundled with premium ChatGPT plans for now.
- Integrations: Dots reaches far more apps (4,000+ vs 220). The video suggests Zapier to fill Grok Bot's gaps (07:07).
- Governance: Before giving any agent logins, check its approval gates and how well it isolates credentials.
Read the full section
- Cost: Grok Bot costs $120–$300 per seat per month. Dots is bundled with premium ChatGPT plans for now. Hermes has no licence fee, but you pay for models and hosting.
- Integrations: Dots reaches far more apps (4,000+ vs 220). The video suggests Zapier to fill Grok Bot's gaps (07:07).
- Governance: Before giving any agent logins, check its approval gates and how well it isolates credentials.
- Selection: The video's practical test is whether you're still using the tool a week later (09:11). That is a reasonable pilot criterion, but it isn't evidence about the tools themselves.
Sources
Read the full section
- Reddit post (feed excerpt)
- YouTube video (auto captions)
- SiliconANGLE: OpenAI launches Dots
- DataCamp: OpenAI Dots explained
- VentureBeat: Grok Bot
- Reworked: xAI launches Grok Bot in beta
- PYMNTS: Grok Bot traction
- xAI docs: Grok 4.7
- Data Studios: Grok 4.7 launch
- GitHub: NousResearch/hermes-agent
- DataCamp: Hermes Agent tutorial
- Chief of Staff agent pattern blog
The source trail.
Sources (12)
Grok Bot vs ChatGPT vs Hermes Agent: 3 Mistakes Nearly Everyone Makes
Feed excerpt only. A transcript of the video linked from this page is supplied (linked video auto captions; https://www.youtube.com/watch?v=FUABuOouS-k); automatic text may contain errors.
reddit.comLinked video: Grok Bot vs ChatGPT vs Hermes Agent: 3 Mistakes Nearly Everyone Makes
linked video auto captions
www.youtube.com