Hermes Agent v0.13.0 adds /goal, Kanban orchestration, and custom LLM providers

• Hermes Agent v0.13.0 shipped with four headline changes, according to Teknium's feature list: Kanban-based multi-agent orchestration, enforced completion via /goal, disk-usage optimizations, and broader extensibility for custom LLM providers and gateway channels.
• The official release post from NousResearch's launch post and the linked GitHub release notes frame the update around speed, reliability, and intelligence.
• Hermes' browser stack is already being used as a self-improving skill loop, where kylejeong's Hacker News example showed a workflow drop from 102 seconds to 35, 23 turns to 8, and $1.46 to $0.28 after two iterations.
• The extensibility push started before the release, with Teknium's plugin post opening plugins for inference providers and gateway channels, while Teknium's blank-profile post added a --no-skills path for starting an agent from scratch.

You can read the full v0.13.0 release notes, browse the upgraded Quickstart docs that Teknium's quickstart post pointed to, and check the browser-harness site behind the browser integration that browser_use's Hermes post showed off. The funnier tell is that Teknium's update meme retweet turned hermes update into a recurring joke before this release even landed.

The biggest product change is that Hermes now treats orchestration and task completion as first-class primitives. Teknium's feature list ties /goal to enforced completion, while the GitHub release notes position Kanban as the control plane for multi-agent work.

That builds on the v0.12 Kanban rollout that WesRoth's Kanban screenshot post described as unlimited boards and projects, plus subscriptions that push project updates into configured home channels. Teknium's Hermes Desktop retweet shows the surrounding desktop tooling was updating alongside Kanban as well.

The more durable shift is how much of Hermes is now becoming pluggable. Teknium's plugin post said plugins can extend both LLM inference providers and gateway channels, and pointed users to docs for Hermes' "total pluggable surface."

At the same time, Teknium's blank-profile post added hermes profile create <name> --no-skills, which strips away built-in skills and makes the agent profile itself the customization boundary. That pairs neatly with the release's custom-provider push from Teknium's feature list.

The clearest performance claim around the release came from the browser stack. In kylejeong's Hacker News example, Hermes used Autobrowse to stop clicking through a task step by step, evaluate JavaScript directly on the page, and save that flow back into a skill.

The reported deltas were unusually clean:

• Time: 102 seconds to 35
• Turns: 23 to 8
• Cost: $1.46 to $0.28

A day earlier, browser_use's Hermes post pitched the same direction more broadly: self-improving browser tools, parallel stealth cloud browsers, and a one-prompt setup via browser-harness.

The early user chatter is less about one killer workflow than about range. NousResearch's analyst retweet highlighted a user running Hermes as an "analyst," while itsPaulAi's Hyperframes post showed a one-line skill install for video generation.

A separate thread from teortaxesTex's stress-test post, teortaxesTex's follow-up, and teortaxesTex's result post described Hermes as "admirably fearless" under messy prompts and concluded, bluntly, that "it works." For an agent framework release, that kind of hands-on vibe check is probably the most valuable datapoint in the whole pile.

One smaller but useful thread around this release window is that Hermes is peeling agent runtime away from simple automation. Teknium's cronjobs post said gateway-connected cron jobs can run scripts, RSS pulls, and diagnostics without keeping an agent in the loop, specifically to cut costs on programmatic tasks.

Meanwhile, Teknium's notification toggle post added a way to disable auto-pushed gateway status and restart notifications when people are serving Hermes agents to customers. That is less flashy than Kanban, but it says a lot about where Hermes thinks deployment is heading.