Video: "New Hermes Agent v0.13 Update just dropped..." by Julian Goldie on YouTube.
What the Tenacity name actually means
Nous Research called this release Tenacity because the central theme is agents finishing what they start. The headline additions — the /goal command and the multi-agent Kanban board — are both designed to solve the same problem: AI agents that begin a job competently and then stop short, lose track of the goal, or need re-prompting to take the next step.
That is genuinely the main failure mode for current agents. Tenacity is a reasonable name for an update that addresses it structurally rather than through better prompting.
What changed in day-to-day use
The most immediate change for existing users is how the agent handles long runs. Before v0.13, if Hermes reached a checkpoint and the session state was large, it would either fail silently or produce inconsistent behaviour on resume. Checkpoints v2 gives you configurable disk limits and proper state pruning — you set a ceiling and Hermes stays inside it. This alone makes the update worth pulling for anyone running Hermes overnight or on shared hardware.
The second change you will notice: messaging integrations (Discord, WhatsApp in particular) have tighter defaults. WhatsApp now rejects messages from contacts not in your allowlist by default. Discord role-allowlists are guild-scoped rather than global. If you had a broad configuration before, some messages that previously reached your agent will stop doing so. Check your integration settings before updating if this matters to your workflow.
What's new that you probably haven't configured yet
The ProviderProfile system is the underreported part of v0.13. It lets you define an AI provider as a plugin file rather than hardcoding it into the Hermes installation. In practical terms: you can now run different providers for different agent tasks in the same setup — GPT-4o for reasoning-heavy work, a fast local model for simpler calls — without branching your config.
There is also native video analysis support via the video_analyze tool, for teams using Gemini or other multimodal models. And xAI Custom Voices is now a TTS option if you want spoken output from your agent. Neither feature requires any extra setup beyond enabling it in config, but they do require provider access.
Before you update: three things to check
First, review your messaging integration allowlists. The new security defaults are better, but they may change who can reach your agent. Second, look at your checkpoint storage path and set a disk limit if you haven't already — the new guardrails do not retroactively clean up old checkpoint data. Third, if you are on an older plugin for a third-party AI provider, check that it is compatible with the ProviderProfile interface — you may need a small update.
None of these are blockers to updating. They are just the places where a default change can produce unexpected behaviour if you are not aware of it.
Where this connects to NordSys
Large releases like v0.13 are exactly the point where an existing Hermes setup can either improve significantly or introduce new issues if the update is applied without review. We help clients set up and maintain Hermes Agent deployments that stay reliable through version changes — right configuration from the start, proper monitoring, and sensible defaults. If your setup is ad hoc and you are not sure where it stands, our AI Agents service is where to begin.
See our AI Agents service →