Ep 881 Research Paper 4:27 w/ Talon & Wildflower

Nous Research Hermes Bot Mode

Hermes Bot Mode makes multi-agent coordination feel like a desktop product by turning existing Hermes profiles into named, persistent bots. Talon sees a real workflow win for small teams, while Wildflower likes the deliberately small technical design but warns that local profiles and inboxes are not a managed enterprise control plane.

Embed this episode

Paste this on any site — the player is a self-contained iframe with no cookies or trackers.

<iframe src="https://sandrise.io/exploring-next/embed/881"
  width="100%" height="180" style="max-width:640px;border:0;border-radius:12px;overflow:hidden"
  title="Exploring Next — Episode 881 audio player"
  loading="lazy" allow="autoplay" referrerpolicy="strict-origin-when-cross-origin"></iframe>
Embed & API docs →
Script GPT-5.6 terra Voice Rime Mist v3

Transcript

Talon The smart part is almost embarrassingly simple. They took profiles people already had, gave them names and faces, and suddenly the agent setup looks like a team instead of a pile of tabs.

Wildflower Exactly.

Talon That matters because the pile-of-tabs experience is where a lot of otherwise decent agent tooling goes to die. You can now point at a researcher, a reviewer, a writer, and see who owns what.

Wildflower And Nous did not build some grand new multi-agent substrate. A Bot Mode bot is just a Hermes profile under tilde slash dot hermes slash profiles slash name, with isolated config, memory, skills, credentials, and chat history. The plugin is basically a humane interface over a primitive Hermes already had.

Talon Right, right.

Wildflower I like that more than I expected to. The storage boundary is legible, and the operations are ordinary profile gateway calls: list, create, describe, configure. That means this is a product layer, not an Agentic Switchboard O S with a twelve-page diagram.

Talon Oh, come ON. You were saving that one.

Wildflower You made us promise not to invent more agent acronyms. I am honoring the treaty.

Talon Okay, that's genuinely funny. But yeah, this is exactly the kind of developer-experience leap I get excited about. A research bot can sit on a reasoning model, while a writing bot uses a cheaper model, and neither drags the other's context around.

Wildflower Mm-hm.

Talon My week has been a little too full of tools claiming they reinvent collaboration, so this one feels refreshing. It just makes the specialist agents visible and gives them somewhere to leave work for each other.

Wildflower The handoff mechanism is refreshingly literal. Each bot has a persistent Agent Inbox, and a message is a real command-line handoff: Hermes runs the target profile, opens that Inbox conversation, and sends the prompt with attribution. The active bot can take an at-mention, hand work over, then report back.

Talon That is the central claim, I think. Coordination does not need mystical swarm behavior. It needs durable identities, separate context, and a receipt that says who handed what to whom.

Wildflower Sure.

Talon Which is very annoyingly consistent with everything we keep finding. The loud feature is multiple bots. The useful bit is that the handoff lands in a persistent, inspectable place.

Wildflower Yeah, although I would not oversell the inspectable part. The article says the desktop tool has no admin console, single sign-on, central audit log, or policy layer. For a solo builder or a small engineering team, that is fine. For a regulated company, an Inbox on individual workstations is not governance.

Talon No way.

Talon Actually, fair. I was about to pitch this as a lightweight team workspace, but it is really a personal or small-team workbench. It can be adopted today, at no license cost, because Hermes Agent and Bot Mode are M I T licensed. That is not the same as being deployable across a company.

Wildflower The rollout itself is interesting evidence of where it fits. It started as a one-day public beta plugin, then got bundled and switched on by default in Hermes Desktop version zero point twenty point three. The standalone Hermes-Bot-Mode repository is archived, and development moved in-tree under the desktop plugin directory.

Talon Okay okay.

Wildflower That speed can mean confidence, or it can mean they are still discovering the sharp edges in public. I could be wrong, but default-on is a bolder product decision than a beta checkbox. At least the group behavior is constrained: two to six bots, up to three serial rounds, and brief replies or passes.

Talon That cap is doing a lot of emotional labor. Every multi-agent demo wants to become a tiny meeting where six models repeat each other until the heat death of the laptop.

Wildflower Stop it.

Talon You know I'm right. Three rounds means it can coordinate a scout, reviewer, and publisher without pretending it solved organizations.

Wildflower And this lands directly in our unresolved harness fight. Editor-native, terminal-boundary, managed runtime, cloud runner... nobody has settled where the agent belongs. Hermes is another workstation-native entrant: the harness lives on the desktop, with profiles as the boundary and the command line as the mailroom.

Talon That makes it more concrete than a lot of the managed-agent pitches, honestly. It is not trying to replace the company. It gives a builder a stable little roster, then lets that roster use the models and tools they already picked.

Wildflower If someone wants to try it, the useful starting points are the hermes-agent repository and the archived Hermes-Bot-Mode repository. I would make two profiles with deliberately different tool access, then test whether the Inbox handoffs preserve enough context to be useful. Skip the pixel pet until after that.

Talon The pixel pet is absolutely going to be the first thing you inspect, Talon. Anyway, this is episode eight eighty-one, and I am pleased our big technical conclusion is: give the bots mailboxes and stop holding meetings with them.