Hermes Agent Update v0.21 is MASSIVE! (Pantheon Release)
The big one: bot mode with persistent context
Hermes v0.21 (Pantheon) bundles six commits from the v0.20 patch. For us, the centerpiece is bot mode, which we already use to research and draft articles on Super Bash. The setup is simple: you create multiple agent profiles, give each one a job, and let them work.
In our repo, three bots run around the clock. One tracks new model releases. One covers other AI tools. One hunts for growth deals, like free credits, reset limits, or discounted plans. They don't commit straight to the main branch. Each bot works on a separate branch, and we review the draft before approving. If we like it, the bot uploads it to the site. So far, that workflow has taken our Hermes agent lessons from 13 tutorials to 24.
The trick is that Hermes doesn't know anything about itself out of the box. You have to give the bots an environment to test in. We point ours at the local setup, documentation, other guides, our own guides, and user input. That is what lets them discover new lessons and write useful material. It is not magic, but it is genuinely useful.
Bot-to-bot DMs, cron jobs, and delegation
The update also adds bot-to-bot DMs. Any Hermes agent can message agents on other profiles or gateways using the peer command. The idea is to let agents work together. We tried it, and the results were messy. There is a lot of drift when they talk to each other. Output gets incoherent. Sometimes one bot does the other bot's job. Our research bot would start writing code, and the coding bot would say, "Looks like it's already done. What do you want me to do next?"
For most users, this feature is unnecessary. If you want agents to collaborate, you will need to supervise closely and explicitly tell them how to communicate. We keep our bots focused on one task each and share repo context instead.
Cron jobs got better too. Before, every run started from zero because there was no persistent memory. A news report on day seven looked nothing like day one. Now cron agents load and update persistent memory like any other agent. There is also a continuity: true setting that carries each run's output into the next, plus a durable notepad per job. That makes scheduled work much more consistent.
Delegation for sub-agents and Kanban workers now has live orchestration. You can list running children, steer one mid-flight with a course correction, or stop it early and keep the partial result. JSON schema validation, per-delegation cost reporting, and a raised default give you a clearer picture of what sub-agent runs actually cost.
Quality-of-life upgrades
MCP servers and catalogs are merged into one desktop page. If you run a lot of MCPs, this is a big win. One command center instead of scattered tabs.
The in-app browser went from guesswork to real interaction. Before, Hermes would guess where UI elements were. Now it can navigate, click, and read directly. That is what lets our bots take screenshots from other websites, and we can watch it happen live on our panel.
There are also six new providers, security hardening, and a CLI glow-up for power users. Overall, v0.21 is a strong update for anyone running Hermes as a 24/7 agent. Just skip the peer-to-peer DMs until they mature.