Verdict: For anyone doing repeated knowledge work — research, drafting, daily briefings, first replies — Hermes Agent's Bot Mode beats a standalone Claude chat tab, because every agent keeps its own persistent memory, runs on a schedule, and can hand work to other agents. Standalone Claude.ai remains better for one-off questions and anything you want zero setup for. The decision is not which model is smarter; it is whether your AI remembers yesterday without you re-pasting it.
TL;DR
Bot Mode ships built-in and default-on in Hermes Agent as of v0.20.3 (released 16 August 2026, GitHub releases).- Each Bot is a full Hermes profile: its own config, memory, skills, and chat history under
`~/.hermes/profiles/<name>/` ([official docs](https://hermes-agent.nousresearch.com/docs/user-guide/bot-mode)).- Group chats let 2–6 Bots deliberate, capped at three serial rounds per message (
[official docs](https://hermes-agent.nousresearch.com/docs/user-guide/bot-mode)).- Bot "Routines" are ordinary cron jobs, so scheduled work still runs when you are not at the keyboard.
-
Standalone Claude wins for ad-hoc questions; Bot Mode wins for anything you do more than twice a week.
-
Last verified: 2026-08-31. Bot Mode is a Hermes Desktop feature that turns Hermes profiles into a roster of named, persistent agents. Each Bot carries its own role (defined in a
SOUL.mdfile), a pinned model of your choice, its own memory, skills, and even a profile picture. Nous Research shipped it bundled and default-on in Hermes Agent v0.20.3 on 16 August 2026 (GitHub releases), and the current patch line has continued at pace — v0.20.5 alone rolled up roughly 323 merged pull requests (GitHub releases).
The engineering decision that matters: a Bot is not a new object type. It is a Hermes profile, with isolated config, memory, credentials, and chat history stored under ~/.hermes/profiles/<name>/ (Hermes profiles documentation). You can open the same agent from the terminal with hermes -p <bot> chat. There are no background daemons and no extra storage beyond what a profile already used.
Claude, Anthropic's assistant, enters the picture as the model you pin a Bot to. Hermes is provider-agnostic — you can run a Bot on Claude, an open-weight model, or a local model — so "Claude + Hermes" means Claude's reasoning wrapped in Hermes's memory and scheduling layer.
The honest comparison is not "Claude vs a different model." It is one chat with amnesia versus a team with a filing cabinet.
| Dimension | Standalone Claude (browser tab) | Hermes Agent Bot Mode |
|---|---|---|
| Memory | Per-chat; you re-explain context in new threads | Persistent per-Bot memory that compounds across sessions |
| Scheduling | None built in; you prompt when you remember | Routines are real cron jobs, visible in hermes cron list |
| Teamwork | One conversation, one assistant at a time | Group chats of 2–6 Bots, up to three serial rounds per message ( docs ) |
| Files and tools | Upload per conversation | Bots read files, run tools, and call skills in their environment |
| Setup cost | Zero | Install + profile creation (minutes per Bot) |
| Cost shape | Flat subscription | Hosting plus per-model API usage of whatever each Bot is pinned to |
| Best for | One-off questions, quick drafts | Recurring workflows: research, first drafts, daily briefings, watchlists |
Numbers and mechanics above are from the official Bot Mode documentation and the NousResearch/hermes-agent releases page.
Memory is the feature that converts an AI tool into an AI colleague. In a browser chat, the fifteenth conversation about your project knows nothing about the first fourteen. In Bot Mode, a research Bot you briefed in August still knows your constraints in December, because its memory lives in its profile directory and is injected into every session (profiles docs).
Practical consequences:
[bot:<name>] <routine>) managed by the same scheduler as every other Hermes job (~/.hermes/profiles/, you can back it up, inspect it, or version it — a different posture from memory held behind a hosted account.
If you want the protocol-level view of how agents exchange context, our guide to the A2A protocol and multi-agent shared memory covers the underlying plumbing. Resist the urge to build a ten-agent empire on day one. The pattern that works is sequential: one Bot doing one job well, then a second once the first is solid.
hermes update (SOUL.md. hermes cron list.
We walk through a larger roster — including model routing and gateway delivery — in How to Build an AI Agent Team With Hermes Agent, and five free Hermes Agent setups you can run today if you want cheaper starting points before paying for API usage.
Cost stacks with parallelism. Each Bot can be pinned to a different model, and three Bots reasoning in a group chat generate three streams of API calls. A roster pinned entirely to a frontier model costs meaningfully more than one subscription chat.
Coordination is capped, deliberately. Group chats settle after three serial rounds, and bot-to-bot messages are picked up on the receiving Bot's next turn rather than interrupting it mid-task (Bot Mode docs). This prevents runaway loops, but it means multi-Bot work is asynchronous — minutes, not milliseconds.
More agents is not automatically better. A stack of poorly-scoped Bots writing into overlapping territory produces noise, not leverage. The discipline — one Bot, one job, then expand — is the actual product.
It adds no new safety model. Every Bot is a standard Hermes agent with the same system access it always had. Supervision and approval habits carry over unchanged.
Q: Is Hermes Agent Bot Mode free?
A: Bot Mode itself is free and MIT-licensed — it ships built into Hermes Agent at no extra charge (GitHub releases). Your real costs are hosting (a small VPS suffices) and the API usage of whichever models you pin each Bot to.
Q: Do I need to know how to code to run multiple Bots?
A: No. Bots are created from the desktop app's Bots tab with a name, role description, and model pick; routines use a structured schedule picker. Terminal access helps for maintenance, but day-to-day operation is clicking and chatting.
Q: Can two Bots share the same memory?
A: By default each Bot is an isolated profile with its own memory under ~/.hermes/profiles/<name>/ (profiles docs). Isolation is deliberate — it keeps a Bot's context clean — and Bots share context with each other through messages and group chats rather than shared memory files.
Q: How is Bot Mode different from just opening several Claude tabs?
A: Tabs have no persistent memory, no schedule, and no way to hand work to each other. Bots keep compounding memory, run routines on a timer, and can deliberate in group chats of 2–6 agents (Bot Mode docs).
Q: Which model should I pin my Bots to?
A: Match the model to the job. Claude is a strong default for writing-heavy and reasoning-heavy Bots; cheaper or local models make sense for high-frequency monitoring routines where per-call cost dominates. You can mix models across the roster.
Q: When did Bot Mode launch, and how do I get it?
A: Bot Mode shipped bundled and default-on in Hermes Agent v0.20.3 on 16 August 2026 (GitHub releases). Existing users receive it via hermes update; new installs from that version onward include it automatically.
Researched and drafted with AI agents; reviewed and fact-checked under human editorial oversight. How we work.