Hermes Agent v0.15: What Changed and Why It Matters

Hermes Agent v0.15 is a large release wave, not a cosmetic version bump. The headline is velocity: faster startup, faster session recall, cleaner core architecture, stronger multi-agent orchestration, better plugin surfaces, and a round of patch fixes that matter for real installs.

Why this release matters#

The best agent releases do not just add buttons. They remove operational drag. Hermes v0.15 does that in three layers:

• Architecture: the core agent loop became easier to maintain and extend.
• Runtime speed: cold-start and per-turn overhead dropped, while session_search became dramatically faster and cheaper.
• Operations: Docker, dashboard, MCP, gateway, skills, and packaging fixes reduce the number of confusing setup failures.

That combination matters because Hermes is not only a chat client. It is a tool-using agent runtime with memory, skills, cron jobs, MCP tools, messaging gateways, and local or hosted deployment choices.

The big architecture win: core agent refactor#

The v0.15.0 release notes describe the central run_agent.py shrinking from more than 16,000 lines to under 4,000, with logic moved into cohesive agent modules. That is not just code cleanup. It makes future fixes safer because tool execution, provider behavior, session handling, prompts, and orchestration are easier to change without touching one giant file.

For users, the practical result is fewer brittle edges over time.

Faster recall with rebuilt session_search#

Hermes memory is strongest when it can recover what happened in older work without asking you to paste context again. v0.15 rebuilt session_search so retrieval is local, fast, and not dependent on a summarizing LLM call. That supports the core Hermes loop: search old sessions, load the right skill, act with tools, verify, and save durable lessons.

Kanban and multi-agent work got more serious#

v0.15 continued the Kanban platform push. Hermes is moving from “one agent in a terminal” toward durable multi-agent task graphs with workers, verifiers, and swarms. If your work is too broad for one chat turn, this is the direction that makes Hermes feel like an AI team instead of a prompt box.

Start with subagent delegation and AI agent cron jobs before trying to automate an entire project board.

Security and secrets improvements#

The release introduced promptware defenses and Bitwarden Secrets Manager support. The theme is clear: as agents gain more tools, credentials and context boundaries matter more. Pair this with MCP security guardrails, profiles, and dashboard access controls instead of connecting every tool to one all-powerful profile.

v0.15.1 and v0.15.2 patch fixes#

The patch releases are worth mentioning because they fixed problems users actually feel:

• Dashboard 401 reload loop in loopback mode.
• Docker --insecure behavior made explicit instead of inferred from bind host.
• Docker MCP command resolution for npx, npm, and node.
• Gateway .md file delivery restored.
• Skills page sidebar and source pills restored.
• Kanban worker termination fixed.
• Packaged plugin manifests included in wheel and sdist.

If you hit install, update, Docker, or dashboard errors around this release, use the install/update troubleshooting guide and the Docker troubleshooting checklist.

Should you upgrade now?#

Yes, if you are on v0.14 or an early v0.15.0 install and you use Docker, dashboard, MCP, gateways, Kanban, Desktop, or packaged plugins. Before upgrading, back up ~/.hermes, check git status if you run from source, and verify with hermes doctor after the update.

If you do not want to own upgrades, Docker, gateway uptime, and dashboard security, use FlyHermes for the managed path and keep self-hosted Hermes for cases where local control matters.