AI LABS walks through the real-world automations they run daily on the Hermes agent, the standout being two cost-control flags that govern when LLM tokens get spent. The Wake Agent flag decides at runtime whether to call the model, firing only when there's a genuine change worth acting on, while the No Agent flag runs jobs with zero LLM involvement but full Hermes context — making most monitoring practically free. On top of that, Hermes' evolving memory and skills let it act as a company "second brain," answer leads over Gmail, track competitors via a PRD skill, and draft social posts for human review.
xurl skill turns long-form video scripts into X and LinkedIn drafts written to a folder for a human review step before anything is posted.The Hermes agent now ships a desktop app that's "basically a wrapper around the agent setup." After testing it against the terminal interface from their previous video, AI LABS found it a more interactive way to work — easier to monitor agents in action and to set up configs. Installation is just downloading the OS installer and running it; the app picks up the configs you already established in the TUI and carries them over. Model, tools, skills, and the memory cap (which budgets memory and profiles by character count) are all managed from the settings and skills panels.
The biggest win for them was profiles. A profile is an individual Hermes agent with its own separate memory and skills, keeping personas isolated. On the terminal you'd run a series of commands to switch between profiles; the desktop app lets you run several in parallel and delegate tasks to each simultaneously.
Wake Agent is the feature that "opens up a lot of ways to use the Hermes agent." It's a flag in the script used inside a cron task that decides, while running, whether the agent should call the LLM at all. If the flag is false, the model never fires for that event — so the agent only acts when there's an actual change worth paying attention to, saving the cost of firing on things it shouldn't handle. On OpenClaw this wasn't possible; the agent ran on every cron job whether needed or not.
Concrete uses shown:
The No Agent flag runs a job without ever invoking the AI model. That sounds like an ordinary cron job, and functionally it is — but it runs inside the Hermes ecosystem, which already holds rich context about how you work, making setup easier than building the same thing elsewhere. Example: a job that monitors metrics like TLS (so the site doesn't silently go down) and Stripe app health, posting health-check updates to Slack. Hermes builds the whole automation from a prompt, pulling in the skills it needs. These jobs are "practically free since there's no agent involved at all," and you can tag Hermes in chat whenever you want it to act on an alert while keeping all existing context.
Connect the Hermes bot to your Slack workspace and let the whole team interact with it. Its evolving memory and skills observe day-to-day tasks and turn them into reusable workflows, accumulating context on what the company is, what it does, and the progress being made. From that it builds a company skill that acts as a complete second brain — a central hub the team can query about anything across the organization. A project manager can use it to track goals, assign tasks to teammates, and update them, because it already holds the context and progress on each initiative. It holds up on long-running tasks thanks to memory editing — the gap with OpenClaw, whose soul file bloats over time and has to be reset to stay effective.
Beyond acting as a second brain, Hermes can take actions on your behalf. It includes built-in skills for Google Workspace: connect Gmail by setting up a Google Cloud project, adding the credentials and callback token (a longer but straightforward setup you can grab from the AI Labs Pro community), and Hermes can read, send, and reply to email. Pair that with a webhook — one app notifying another that an event happened — and Hermes can monitor incoming mail, spot likely leads, and auto-reply using everything it knows about the company. If a meeting is involved, it can even schedule it on your calendar around your open slots.
As covered in a prior video, keeping your PRD as a skill (rather than a plain document) matters because a skill loads only when needed and stays in the fresh part of the context window where the model is paying attention — so the agent never loses the requirements that matter. Since the skill already knows the product, you can reuse it to analyze competitors: a cron job (typically weekly, matching how often competitors ship) studies them, updates the PRD with potential features to add, and maintains a separate file tracking competitors' progress. Those files become part of the agent's context, so you can act on its recommendations as you build.
An xurl skill automates posting to X. AI LABS use it to repurpose content — turning long-form video scripts into X and LinkedIn posts. Because Hermes already knows their voice and writing style, a custom skill takes a video script and writes drafts for both platforms into a specific folder rather than posting directly, preserving a human review step (the X post respects the character limit). After review, you ask Hermes to publish and it invokes the skill to post.
"The Hermes agent is one of the most powerful agents out there, but people don't really know how powerful it can get when it's actually used to optimize the processes they interact with daily."
"If the wake agent flag is set to false, the LLM won't get called for that event. So, the agent only fires when there's an actual change worth paying attention to."
"The big upside of these jobs is they're practically free since there's no agent involved at all."
"An agent like Open Claw can't keep this up because its soul file gets so bloated that it starts losing sight of what it actually needs to focus on. So you'd have to reset that soul file."
"A skill only gets loaded when it's actually needed and stays in the fresh part of the context window where the model is paying attention, so the agent never loses track of the requirements that matter."
| Time | Topic |
|---|---|
| 00:00 | Intro |
| 00:19 | The new Hermes desktop app |
| 01:56 | The Wake Agent feature |
| 03:39 | The No Agent flag |
| 04:54 | Hermes as your company's second brain |
| 06:34 | Letting Hermes take actions |
| 08:15 | Competitor analysis with PRD skills |
| 09:29 | Automating your social media |
| 10:38 | Sponsor |
xurl skill, TLS & Stripe health checks