Hermes v0.21 — the Pantheon Release — is out as of 31 August 2026, and it turns Hermes from a solo assistant into a team of persistent, named agents that talk to each other. It shipped from Nous Research under the full version number Hermes Agent v0.21.0, announced by the official Nous Research account and by Teknium, Nous Research's cofounder and lead engineer on Hermes Agent. The headline, straight from the announcement thread: Hermes is now a system of persistent, named agents — each with a face, a role, memory and skills — that you can group into chats or at-mention for collaboration.
📺 Watch: Hermes Bot Mode DESTROYS Grok Bot? (FREE!)
🔥 Get the Agent OS as a free bonus: AI Profit Boardroom members get the full Agent OS zip, prompt libraries, daily tutorials and weekly live coaching calls. → Get inside
I'm Julian Goldie. I run AI agents all day across my SEO business and a YouTube channel with 400K+ subscribers, and I've covered every Hermes release since bot mode first landed. This one is hours old, so I'm sticking to what's officially confirmed, what early community coverage adds, and what I'd actually do with it today.
What the Pantheon Release Actually Is
The name is doing real work. A pantheon is a roster of named characters, and per the announcement that's exactly what Hermes becomes in this release: not one assistant you prompt, but a set of persistent agents you hire. Each bot keeps its own face, role, memory and skills, and you can drop them into group chats together or at-mention one for a specific handoff — they collaborate rather than queue.
The scale behind it is serious too. Early coverage puts the release at around 5,800 commits from more than 760 contributors, which is not a point release — that's a platform push. And if you've read my Hermes 3D office piece, you know the roster already had a visible workplace; the Pantheon Release is the moment the characters in that office start properly working with each other instead of just sitting near each other.
Hermes v0.21 Changelog: The Official Features, Explained
Teknium's post lists five headline items — Bots Mode, Agent-to-Agent Comms, Persistent Multi-Gateway Connections, Subagent Steering and Expanded Connectors Access — plus "a lot more". Here's each one in plain English, with what it means in practice.
Bots Mode
Bot mode itself arrived in the last version, and my Hermes bot mode guide covers the foundations: a roster of named bots, each with its own role, model, memory and skills, living in group rooms with at-mention handoffs. In v0.21 that roster becomes the centre of the product rather than a feature on the side. The word that matters in the announcement is persistent — your bots stay who they are, remember what they have done, and are still there tomorrow.
Building your own bot team this week? I've documented my exact setup, rosters and workflows — grab my Agent OS build-outs and bot-mode tutorials inside AI Profit Boardroom and copy what already works instead of starting from scratch.
Agent-to-Agent Comms
This is the one I care about most. Per early coverage of the release notes, bots can now DM each other and get real replies back — agent-to-agent communication as a first-class feature, not a workaround. In practice, your research bot can ping your writing bot with findings and get an answer, without you playing messenger in the middle. I already rate Hermes as the best messaging-app agent going; v0.21 makes the messaging run between the bots themselves.
Subagent Steering
Until now, handing work to a subagent was mostly fire-and-wait: it went off, did the task, and you found out at the end whether it had drifted. Per the community-confirmed details, v0.21 lets you steer a subagent live mid-task — nudge it, correct it, redirect it while it is still working. For long-running jobs, that's the difference between delegation you trust and delegation you babysit after the fact.
Persistent Multi-Gateway Connections
Conceptually, your agent can now hold more gateway connections open at once, and hold them persistently, so an always-on setup stays plugged into multiple channels at the same time rather than juggling one. The same persistence theme shows up in scheduling: per the coverage, cron jobs now keep continuity, meaning a scheduled job remembers what it did yesterday instead of waking up with amnesia. If you run daily automations like the ones in my Hermes loop setup, that continuity is quietly one of the biggest wins in the whole release.
Expanded Connectors Access — Plus the Browser Driver
Wider connectors access means your bots can reach more of your actual tools, which is what turns a clever chat toy into staff. And per early coverage, the agent also gains an in-app browser driver, so it can drive the web itself rather than only read what you paste in. Put those two together with agent-to-agent comms and you can see the shape of it: a team that reaches your stack and the open web without you as the bottleneck.
📺 Watch: Hermes3D Is INSANE: AI Agents Now Have a 3D Office
Hermes v0.21 vs v0.20: The Honest Delta
Let me be straight, because half the hype threads won't be. Hermes v0.20 already gave bots names, roles, models, memories and group rooms — I covered it at the time, and it was genuinely good. What the Pantheon Release adds is persistence and conversation: bot-to-bot DMs with real replies, agent-to-agent comms promoted to a first-class feature, live steering of a subagent instead of waiting for it to finish, cron jobs that carry yesterday's context forward, more gateway connections held at once, and wider connector access. Short version: v0.20 hired the team, v0.21 got the team talking to each other.
Why the Pantheon Release Matters
My whole Agent OS thesis has been the same for a year: mission control, shared memory, orchestration — one place where a team of agents is directed, remembers what matters, and hands work around. Releases like this keep validating that team-of-agents pattern, and this is the loudest yes so far. A persistent roster only earns its keep with a proper command centre over it and a memory OS underneath it, and v0.21 pushes both from optional to essential.
The community reaction points the same way. Early users describe the release as evolving Hermes from a solo assistant into an always-on team that delegates smoothly, and one agency operator on X put it bluntly: this puts serious pressure on rival agent bots, with the comparison chatter naming Grok Bot and OpenClaw as the ones feeling the squeeze. I'm not declaring winners on day one, though. Which brains I actually run the new bots on will come down to my usual Goldie Bench testing, same as every release before it.
📺 Watch: Hermes AI Agents Just Went Portable
How to Get Started Today
The release is hours old, so keep it small and conceptual — and check the official release notes for exact steps, because I'm not going to invent menu paths that may have shifted by the time you read this. Here's the play I'd run with v0.21 today:
- Update Hermes first and confirm your existing setup survived the jump before touching anything new.
- Start with two named bots and one shared room — a researcher and a writer, say — rather than a ten-bot empire on day one.
- Give them real jobs from your business, then let them DM each other to hand work across instead of relaying everything yourself.
- Add persistence last: one scheduled job with continuity, steer it live when it drifts, and only scale the roster once those two bots have earned it.
Pantheon Release Features at a Glance
| Feature | What it does | Who it's for |
|---|---|---|
| Bots Mode | Persistent named agents with faces, roles, memory and skills | Anyone building a repeatable AI team |
| Agent-to-Agent Comms | Bots DM each other and get real replies back | Multi-step workflows without a human relay |
| Subagent Steering | Redirect a working agent live, mid-task | Long tasks that tend to drift |
| Persistent Multi-Gateway Connections | More channels held open at once, persistently | Always-on, multi-platform setups |
| Cron continuity | Scheduled jobs remember what they did yesterday | Daily reporting and recurring ops |
| Expanded Connectors + browser driver | Wider tool access, and the agent drives the web in-app | Research, publishing and outreach workflows |
FAQ: Quick Answers on the Pantheon Release
What is Hermes v0.21?
It's the newest version of Hermes Agent — v0.21.0, nicknamed the Pantheon Release — shipped by Nous Research on 31 August 2026. It turns Hermes into a system of persistent, named agents that hold their own roles, memories and skills, and collaborate with each other in shared chats.
What is the Pantheon Release?
It's the name Nous Research gave this update, announced by the company's account and by cofounder Teknium. The pantheon idea is the point: a roster of distinct characters rather than one assistant, each with a face and a job, grouped into chats or at-mentioned when you need a specific one.
When did the Pantheon Release come out?
31 August 2026 — the same day I'm writing this. The announcement came from the official Nous Research account alongside Teknium's changelog post, and early community coverage of the release notes filled in the practical details within hours.
Is this still the same Hermes project as before?
Yes — it's the same Hermes Agent project from Nous Research shipping a new version, not a separate product. The announcement focused on features rather than any change to how the project is offered, so on anything beyond that, treat the official release notes as the source of truth.
Do I need bot mode experience before trying this?
No, though it helps. This release builds directly on the bot mode that v0.20 introduced, so if you've already run a roster you'll feel at home immediately. If you haven't, start with two bots and one room as above — the learning curve is honestly gentler than running one overloaded assistant.
My Verdict on the Pantheon Release
Hermes v0.21 is the clearest signal yet that the solo-assistant era is ending. Named bots that persist, message each other, take live steering and keep continuity across scheduled runs — backed by roughly 5,800 commits from over 760 contributors — is a team, not a feature. I'll be testing it all week on real client work, and my honest day-one read is simple: if you've been waiting for the right moment to build your agent team, this release just made the decision for you.
Want this running in your business, not just in your bookmarks? Join AI Profit Boardroom and get my Agent OS build-outs plus step-by-step bot-mode tutorials — I share every roster, prompt and workflow as I stress-test releases exactly like this one.
