> ## Content Index
> Fetch the complete content index at: https://globalfeed.ai/llms.txt
> Use this file to discover other available public pages before exploring further.

# An agent society from Nous Research: in Hermes, bots now DM each other
- URL: https://globalfeed.ai/en/an-agent-society-from-nous-research-in-hermes-bots-now-dm-each-other/
- Published: 2026-08-31T21:53:16.000Z
- Updated: 2026-08-31T21:53:16.000Z
- Description: The Pantheon release of the open-source Hermes Agent turns a single-agent tool into a team: bots with identities in group chat, durable agent-to-agent messaging, scheduled jobs with memory, and live subagent steering.
- Author: GlobalFeed Editor
- Tags: nous, hermes, agents, open-source, x-NousResearch, x-Teknium, dil-en, elle, video

Nous Research, open-source AI's most liked lab, has shipped the [Pantheon release](https://github.com/NousResearch/hermes-agent/releases/tag/v2026.8.31?ref=globalfeed.ai) of its Hermes Agent framework. Running on desktop (macOS, Windows, Linux) and the command line, the framework shifts gears with this version: no longer a single-agent tool but a small society of agents.

Two ideas sit at the center. First, Bot Mode: multiple agents live in the same group chat with their own identities and avatars; you hand out work with @-mentions, managing a team rather than a tool. Second, agent-to-agent communication: with the new "hermes peer" command, bots DM each other directly across machines and gateways. You tell your research agent to hand its findings to your coding agent, and the exchange does not evaporate; it lands durably in each agent's chat where you can read it.

The rest of the changelog is dense too: scheduled jobs now keep memory between runs (a monitoring agent remembers its past reports and stops double-alerting); delegated subagent work can be steered live, stopped early with results retained, and validated against JSON schemas; MCP connections moved into a single command center with drag-and-drop imports and health checks; six new model providers were added. The release consolidates roughly 5,800 commits and 2,475 merged pull requests.

The bigger picture: the next question in the agent world is not what a single agent can do but how agents work together. While closed labs bury that layer inside their products, Nous has put the same layer on the table as open source; you can wire it into Slack, Telegram and your existing channels and stand up your own agent team today.