Hermes v2026.5.28 Update: Hermes Agent v0.15.0 (2026.5.28) — The Velocity Release

Hermes Agent v0.15.0 is all about speed — starting faster, running faster, shipping work faster, and making it easier for the community to contribute code.

What’s New

The Big Refactor (finally!)
The heart of Hermes — the run_agent.py file — was a 16,083-line monster that took 90 seconds to open in an editor. It’s now been split into 14 clean modules under agent/, shrinking to just 3,821 lines (-76%). Think of it like taking a giant tangled ball of yarn and winding it into 14 neat skeins. Nothing changed in behavior, but now plugin authors can actually find what they’re looking for, and future development moves much faster.

Kanban grows up into a real multi-agent platform
What used to be a simple task board is now a full orchestration system. You can:

Auto-decompose one big task into a tree of sub-tasks (like breaking a novel into chapters, then paragraphs)
Create a full swarm topology with one command — root, parallel workers, a verifier, and a synthesizer all connected via a shared blackboard
Assign different AI models to different tasks — use a cheap model for boilerplate, a powerful one for hard problems
Schedule tasks to start at specific times, set per-task worktrees and branches, and even configure claim timeouts so stuck tasks don’t block the board

Cold-start performance wave continues
Another second shaved off launch time. 47% fewer function calls per conversation. And hermes --version now beats Codex CLI in head-to-head benchmarks. It’s like upgrading from a bicycle to an electric scooter — same route, way less effort.

Session search is now 4,500× faster and free
Searching through past conversations used to be painfully slow. Now it’s nearly instant, and you don’t need a paid API key to use it.

Security: Promptware defense + Bitwarden Secrets Manager
Hermes now defends against “Brainworm-class” attacks — malicious prompts that try to hijack the agent. And instead of managing API keys for every provider, you can now use one Bitwarden bootstrap token to unlock all your secrets. Less key management, fewer leaks.

Skill bundles
One slash command can now load an entire workflow — like a recipe card that includes all the ingredients and steps, not just a single instruction.

Improvements

Ink TUI gets a multi-session orchestrator — manage several agent conversations from one screen
Two new image generation providers (Krea 2 Medium + Large) and FAL is now a plugin
Nous-approved MCP catalog with an interactive picker — browse and add tools without leaving the terminal
23rd messaging platform added: ntfy (push notifications to your phone)
Deeper xAI integration — Web Search plugin, OAuth proxy, auto-detection of retired models with hermes migrate xai, natural speech-tag pauses in TTS, and a base_url leak guard

Fixes

15 P0 (critical) and 65 P1 (high-priority) issues closed, plus 19 security-tagged fixes. Notable: stale-task detection, respawn guards, and retry fingerprinting in Kanban; a drag-to-delete trash zone for tasks; and a fix for workers reporting through /workers/active and /inspect endpoints.

What This Means for You

If you’re a daily Hermes user, you’ll notice the faster startup immediately, and session search will feel like magic. If you’re a developer, the refactored codebase means you can finally contribute without a machete. And if you’re running multi-agent workflows, the new Kanban features turn Hermes from a solo tool into a small orchestra you can conduct from your terminal.

The velocity release isn’t just about speed — it’s about removing friction everywhere: from launch, to coding, to orchestrating complex tasks.

📎 Official release notes: https://github.com/NousResearch/hermes-agent/releases