Hermes Agent Bot Mode launch

Strip away the avatars and the roster view, and a 'bot' in Bot Mode is just a Hermes profile: isolated config, memory, skills, credentials, and chat history sitting under ~/.hermes/profiles/<name>/ on disk. Nous Research's own repository is explicit about how little was actually built - no core patches, no background daemons, no extra storage layer, everything is standard Hermes surface ^[1]. What Bot Mode adds is naming, identity (a role, a picture, a description), and a visual switcher that replaces the old single-agent session list.

That matters because Hermes Agent already shipped the underlying primitive - 'Profiles,' for running isolated agent instances - back in its v0.6.0 release, roughly four and a half months before Bot Mode existed as a concept. Building on top of an existing, battle-tested primitive instead of inventing a new multi-agent subsystem is a low-risk way to ship a headline feature fast, which is exactly what happened: from one-day public beta to bundled-and-default-on took two days. It also explains the most common pushback the feature gets from its own users: it is, structurally, a skin.

Bot Mode's public debut landed within days of xAI shipping its own 'Grok Bot' feature, and coverage of the Nous Research launch explicitly framed it that way ^[2]. When a user drew the parallel directly to Teknium, the Nous Research co-founder who built Bot Mode, his reply was two words: 'That's the point.' That is about as close to an on-record admission of competitive intent as a launch gets.

The substance of the comparison is licensing, not raw capability - by most accounts Grok Bot still has an edge in per-bot cloud and browser compute. What Nous Research is selling instead is openness: Bot Mode and the Hermes Agent it runs on are both MIT-licensed, meaning anyone can inspect, fork, or redistribute the multi-bot tooling, whereas Grok Bot is closed inside xAI's product. That framing targets a specific audience - individual developers and small teams who want to own their agent stack - rather than the enterprise infrastructure buyers a closed competitor might chase.

The feature everyone is sharing clips of is inter-bot messaging, and mechanically it is simpler than it looks: bots communicate through a persistent Agent Inbox, and @mentioning a bot triggers a real CLI handoff rather than routing through some new messaging layer ^[1]. The bot that made the handoff waits for a reply and folds the result back into its own conversation, so from the user's seat it reads like a specialist colleague being looped in rather than a separate program being invoked.

That handoff model scales to group chats of two to six bots, with up to three serial rounds of responses allowed per user message before the conversation returns control to the human ^[2]. The tradeoff Nous Research's own documentation flags is that delivery between bots happens per-invocation, not through an instant or interrupt-driven channel - fine for sequential research-then-review workflows, but a real ceiling for anyone trying to build tightly coupled, low-latency multi-agent pipelines on top of it.

The pace of shipping is itself notable: a one-day public beta on August 14, folded into a bundled, default-on release two days later (v0.20.3, roughly 125 pull requests), followed by a patch release two days after that (v0.20.4, roughly 74 pull requests) fixing group-chat bugs and adding a tabbed sessions/bots sidebar ^[3]. That is a company iterating on a headline feature in near real time based on how early users are actually breaking it.

Early reaction to what they built is genuinely split. One camp - visible in a Reddit build log where a tester assembled a three-bot stock-research team of Researcher, Risk Analyst, and Thesis Editor - treats reusable, narrowly scoped specialist bots as a real productivity gain, cutting the per-turn token cost of loading tools into one do-everything profile. The other camp calls it a Hermes Desktop front-end novelty: the multi-agent capability 'always existed' via Telegram or Discord bot-to-bot setups or the native Agent-to-Agent protocol, and Bot Mode just puts a face on it. A practical wrinkle sits underneath both views - shared knowledge duplicated across bots tends to drift out of sync, which is why testers recommend keeping it in a single working-directory-level file rather than copying it into every bot; Nous Research's team has said a proper shared-memory system across profiles is planned for a future release.