Research Brief: Nous Research Ships Bot Mode for Hermes Agent (v0.20.3)

• Source: MarkTechPost (https://www.marktechpost.com/2026/08/17/nous-research-hermes-bot-mode/)
• Author: Michal Sutter
• Date Published: 2026-08-17 / 2026-08-18
• Target File: ~/topics/research/nous-research-hermes-bot-mode.md
• Subject: Hermes Agent Bot Mode Architecture, Multi-Bot Handoffs, and Profile Orchestration

1. Executive Summary

Nous Research has introduced Bot Mode into Hermes Agent (bundled default-on as of desktop v0.20.3 via PR #87886). Bot Mode replaces single-agent session views with an interactive roster of named, specialized bots.

Rather than inventing a new orchestration runtime or database abstraction, Bot Mode leverages Hermes' native profile filesystem (~/.hermes/profiles//). Each bot operates with its own isolated memory, skills, toolset configuration, model pinning, and credentials. Communication occurs via standard CLI invocations into persistent Agent Inboxes, enabling seamless @mention delegations and multi-bot group chats.

2. Under-the-Hood Architecture

Profile-Centric Storage Model

• Filesystem Isolation: Every bot corresponds 1:1 to a directory under ~/.hermes/profiles//.
• Zero Overhead: No new backend database layer is introduced; profiles store their own config.yaml, SOUL.md (persona/protocol), MEMORY.md, credentials, and local skill sets.
• Gateway RPC Integration: Roster management operates over existing profiles.* gateway methods (list, create, describe, configure).

Inter-Bot Messaging & Handoff Protocol

• Persistent Agent Inbox: Each profile maintains an internal conversation named Agent Inbox.
• CLI Handoff Mechanism: When an agent delegates or mentions another bot, Hermes triggers an internal CLI call:

`bash

hermes -p chat -c "Agent Inbox" -q ""

`

• Attribution & Reply Rules: The target bot's SOUL.md instructs it on parsing arriving tasks, processing the prompt with full profile-level tool access, and delivering structured results back to the caller.

Multi-Bot Group Chats

• Capacity: Supports shared multi-agent rooms with 2 to 6 bots.
• Execution Lifecycle: A user message initiates up to 3 serial rounds of member turns.
• Addressing: Explicit @mention targets specific bots; unmentioned prompts trigger all room members to respond concisely or yield (pass).

Scheduled Routines

• Bot-specific automated tasks are mapped directly to standard Hermes cron jobs with the namespace prefix:

[bot:]

• Viewable and manageable via standard hermes cron list.

3. Practical Implications & Use Cases

1. Tiered Compute & Cost Optimization:

- Assign specialized bots to disparate models based on task complexity (e.g., pinning a lead researcher to a heavy reasoning model while pairing with a lighter local quant model for formatting or summary extraction).

2. Strict Separation of Concerns:

- Prevents prompt pollution and cross-domain memory leakage across distinct business, engineering, and personal domains.

3. Automated Pipeline Handoffs:

- Multi-stage workflows where an intake bot routes tasks to an execution specialist via @bot_name without requiring external workflow wrappers.

4. Operational Considerations

• Enterprise Governance: Designed as a sovereign, workstation-level developer tool (MIT licensed). Lacks centralized enterprise SSO, administrative tenancy, or centralized audit logging.
• Codebase Location: Integrated in-tree within apps/desktop/src/plugins/hermes-bots/ following the migration from Teknium's initial public beta plugin.