Video: "Hermes Agent Update v0.21 is MASSIVE! (Pantheon Release)" by Julian Goldie on YouTube.

What Bot Mode means now it runs everywhere

In v0.20.3, Bot Mode was limited to the Hermes Desktop application. You could create named Bots with their own model, memory, and skills, and those Bots could message each other inside a shared collaboration room on your local machine. The limitation was exactly that - local. The moment a Bot needed to do something outside the desktop app, you were back to manual coordination.

Pantheon removes that boundary. Bot Mode is now part of the full Hermes stack, including cloud-deployed agents. A Bot running in Hermes Cloud can send and receive messages from another Bot running locally or in a different cloud container. The collaboration room concept carries over: two to six Bots share a structured message thread and divide work between them without human input at each step. The setup is the same as before - name the Bot, pick its model, write its system prompt, connect its tools - but the scope of what those Bots can reach is considerably wider.

Memory-backed cron jobs: scheduled tasks that know what they did before

Cron jobs existed in Hermes before Pantheon, but each run started cold. If you scheduled a daily content review, the agent that ran on Tuesday had no record of what the Monday run found. You could work around this by writing outputs to a file and reading it back at the start of each run, but that required you to build the persistence yourself.

Pantheon adds a memory layer directly to scheduled jobs. A cron task can now read its own run history before it begins and write a structured summary when it finishes. In practice this means a monitoring agent can say "this metric dropped two days in a row" rather than only "this metric is low today". A content agent can avoid repeating topics it covered last week. The memory is scoped to the individual cron job, so there is no bleed between unrelated scheduled tasks.

Live subagent steering

When Hermes spins up a child agent to handle part of a larger task, previous versions gave you two options once it was running: wait for it to finish or cancel the whole job. Pantheon adds a third option - intervene without stopping.

From the Hermes interface you can now list every active child agent and see what each one is currently doing. If a child agent has gone down an unproductive path, you can redirect it - send a new instruction that changes its focus without resetting it from the beginning. If a child agent is clearly stuck or producing bad output, you can stop just that one while the rest of the workflow continues. This matters most in multi-agent pipelines where one step blocking would previously freeze everything downstream of it.

The MCP command centre and other changes

The MCP command centre - the interface for managing Model Context Protocol connections - was rebuilt in Pantheon. The previous version required you to edit a configuration file to add, remove, or adjust MCP tools. The new interface handles this from a panel inside Hermes, with status indicators showing which connections are live and which are failing. Browser control inside the desktop app was also extended, with more reliable page interaction for agents running web-based tasks. Pantheon also added new model providers and tightened the security rules around what agents can do without an explicit confirmation step.

What to do with this if you are already running Hermes

The memory-backed cron jobs are the change most immediately useful to existing users. If you have any scheduled monitoring or content tasks running on a daily cadence, enabling memory on those jobs takes a few minutes and removes the most common reason to abandon the approach - the agent not knowing what happened yesterday. Subagent steering is worth setting up if you run multi-step pipelines; the ability to redirect rather than restart will save time on complex jobs that occasionally go sideways.

Bot Mode across the full stack matters if you want to run a workflow that splits across cloud and local agents, or if you want two Hermes Cloud instances to coordinate directly. For most small-business setups, the practical change is that collaboration rooms are no longer a desktop-only feature.

Where this connects to NordSys

Configuring Hermes Bot Mode, writing the system prompts for each named Bot, setting up memory-backed cron jobs, and connecting MCP tools takes time if you have not done it before. Our AI Agents service covers this end to end - we design the workflow, set up the agents, and test the pipeline before handing it over.

See our AI Agents service →