Automate your marketing with Wonda inside Hermes Agent

Most AI marketing stacks are fake automation.

They give you five tools, a scheduler, a spreadsheet, and a new promise that this time the workflow will be "agentic." Then you still end up operating the whole machine yourself.

If your marketing automation still requires you to manually operate five tools, it is not automation.

The useful version is simpler: Hermes is the interface, and Wonda is the engine Hermes already knows how to use. You talk to Hermes. Hermes runs the content workflow. Wonda handles the execution behind the scenes.

That is the point of this setup. You do not install Hermes so you can learn another CLI. You install it so your content pipeline starts to feel like one system instead of a pile of moving parts.

If you want the editor-first version of the same idea, read You Don't Need to Learn the CLI: Let Claude Code Run Wonda for You. This article is the daemon-first version: always-on, message-driven, and good at keeping a content loop moving while you are away from the keyboard.

Key Takeaways

• Hermes Agent is Nous Research's self-hosted runtime for always-on agent workflows, with skills loaded from ~/.hermes/skills/ and skills.sh.
• Wonda already ships there as a published skill at skills.sh/degausai/wonda.
• The useful abstraction is not "learn Wonda commands." It is "use Hermes as the single tool for your marketing pipeline."
• The Hermes-native install path is one command: hermes skills install degausai/wonda/skills/wonda-cli.

TL;DR

Install Hermes. Add the Wonda skill with hermes skills install degausai/wonda/skills/wonda-cli. Connect your accounts. After that, treat Hermes as the only surface you talk to. Ask it to research competitors, draft posts, generate assets, review a weekly batch, and publish or schedule the approved pieces. Wonda is still doing the work, but you should rarely need to think about Wonda directly.

Verification note

This article was checked against the current Hermes installation, messaging, skills, and cron docs, plus Wonda's public skill and MCP surfaces, on April 23, 2026. The easiest Hermes setup remains the Wonda skill; the public MCP server is real, but better treated as an optional alternative than the main path.

What is Hermes Agent, practically?

Hermes Agent is an open-source, self-hosted personal agent runtime from Nous Research, distributed under MIT and installable with a single command. It runs as a background daemon, keeps its own SQLite memory, and can talk to you through a local CLI or messaging gateways like Telegram, Discord, Slack, and WhatsApp (Hermes architecture).

That matters because marketing work is not a single prompt. It is a loop. You need the agent to remember your brand, carry forward what performed last week, stay available when a scheduled job fires, and route results back to the same place you already check messages.

Hermes is good at that shape of work. It is not primarily an editor companion. It is a persistent operator.

The core stats are strong enough to make the case on their own: Hermes supports 18+ LLM providers, runs 47+ built-in tools across 19 toolsets, and supports multiple terminal backends through the same runtime (Hermes architecture).

Citation capsule. Hermes Agent is an MIT-licensed, self-hosted personal agent runtime from Nous Research. It runs as a daemon, stores memory locally, and supports 18+ LLM providers plus 47+ built-in tools across 19 toolsets according to the Hermes architecture docs.

Why does Hermes fit marketing operations so well?

Because the work is operational, not purely creative.

The hard part is rarely "write one caption." The hard part is keeping a whole system moving: gather context, draft ideas, generate assets, reject weak outputs, schedule the approved work, then use analytics to shape the next cycle.

That is exactly where a daemon is better than a one-off interactive session. Hermes can stay alive on a server, receive instructions from Telegram, run recurring jobs through its scheduler, and return the results to the same chat. The user experience becomes much simpler:

1. you ask Hermes for the outcome
2. Hermes runs the content workflow
3. you approve what matters
4. the loop keeps going

The important thing is that the user stays at the level of intent. The agent handles the mechanics.

Why do Hermes skills make this simple?

Hermes skills are Markdown procedures the agent can load on demand. They live in ~/.hermes/skills/, and published skills can be installed from skills.sh with a single command.

That is why Wonda fits Hermes so cleanly. You are not writing a custom integration. You are not building a plugin from scratch. You are installing an already-published skill that tells Hermes how to operate the existing Wonda surface.

For this article, the important consequence is not technical elegance. It is simplicity:

• Hermes stays your interface
• Wonda stays the execution layer
• you do not need to learn the implementation details first

This is the same reason good software feels simple. You should not need to understand the lower layer to use the product at the upper layer. Hermes skills create that separation for agent workflows.

Citation capsule. Hermes skills are auto-discovered from ~/.hermes/skills/ and can be installed from the public skills.sh registry. That makes a published skill like Wonda feel native inside Hermes: install once, then use Hermes as the interface.

What does Wonda actually add to Hermes?

Wonda gives Hermes a ready-made marketing capability layer.

After installation, Hermes can use Wonda to handle the core parts of a modern content pipeline:

• research competitors and account context
• generate images, video, captions, and related assets
• edit and format media for social use
• publish or schedule across supported channels
• pull analytics back into the next content cycle

That is the useful framing. Not "now Hermes knows command X and flag Y." The point is that Hermes can now act like a marketing operator instead of a general-purpose assistant with no hands.

And this is already a real install path, not a custom integration project. Wonda is published at skills.sh/degausai/wonda, and it also appears in the community-curated awesome-agent-skills list under Marketing. The practical takeaway is simple: you are installing an existing skill, not wiring Hermes to marketing APIs yourself.

How do you get Wonda working inside Hermes quickly?

The fast path is still the skill path.

Step 1: install Hermes

curl -fsSL https://raw.githubusercontent.com/NousResearch/hermes-agent/main/scripts/install.sh | bash

Hermes's own install guide covers the rest of the environment details, including messaging setup and provider configuration (installation docs).

Step 2: add the Wonda skill

hermes skills install degausai/wonda/skills/wonda-cli

That is the key step. Once the skill is present, Hermes can discover it and use it as part of normal work.

Step 3: authenticate Wonda

wonda auth login

If you are on a headless host with no browser, use WONDA_API_KEY instead. That avoids the first setup trap most VPS users hit.

At that point, the important part is over. The stack is installed. From there, the real discipline is simple: stay at the Hermes layer unless you have a specific reason not to.

What should the experience feel like?

It should feel like you have one marketing operator, not five disconnected tools.

You should be able to tell Hermes things like:

• "Plan next week's content for our product launch"
• "Review last week's Instagram performance and adjust the next batch"
• "Draft three LinkedIn posts and two Reels from this product update"
• "Leave anything risky as a draft and send me the best two options"
• "Schedule the approved batch and give me a short summary in Telegram"

That is the mental model that matters. Hermes is where you make requests, review outputs, run recurring jobs, and keep the feedback loop alive. Wonda stays underneath that experience instead of becoming the thing you have to think about all day.

What does a first useful workflow look like?

The best first workflow is not "run a long chain of commands." It is "send Hermes one request that touches the whole loop."

For example, once your Telegram gateway is live, a first useful instruction is:

Plan three posts for this week's coffee launch.
Use our brand voice, review last week's performance,
and leave anything weak as a draft for approval.

That one request is enough to prove whether the setup is valuable. If the system comes back with a coherent mini-batch, clear rationale, and a manageable approval surface, the pairing is working.

The point is not that Hermes magically does marketing. The point is that Hermes can coordinate the marketing workflow through Wonda without forcing you to micromanage the machinery.

Minimal Telegram setup

If you want the messaging path, Hermes's shortest supported setup is:

hermes gateway setup

If you prefer manual config, Hermes's Telegram docs show the minimum TELEGRAM_BOT_TOKEN and TELEGRAM_ALLOWED_USERS setup in ~/.hermes/.env plus normal gateway start commands (Telegram docs).

That is enough to make Hermes usable as a real operating surface instead of a terminal-only daemon.

How should weekly automation work?

The clean way to think about weekly automation is not "teach Hermes a hundred steps." It is "give Hermes a standing responsibility."

Hermes already has a scheduler. The gateway daemon runs cron jobs every 60 seconds, starts fresh agent sessions for due jobs, and can attach skills directly to those jobs (cron docs).

So the right weekly pattern is:

1. keep Hermes running through the gateway
2. create a recurring cron job
3. attach the Wonda skill
4. give the job a short, high-level prompt
5. review the output where it is delivered

That prompt should describe outcomes, not implementation details. For example:

Every Monday at 9am, prepare this week's content batch for our coffee brand.
Review last week's performance, draft the strongest 5 ideas, generate the top 3,
leave risky items as drafts, and send me a short summary of what is ready to ship.

That is enough. Hermes can carry the task. You do not need a long verb-by-verb tutorial for the setup to become useful.

Where should the human stay in the loop?

At approval and taste.

That matters more than whether the underlying system can technically publish on its own. The safest pattern is still:

• let Hermes prepare the batch
• let Wonda do the heavy lifting underneath
• keep human approval at the publish edge

That is especially true on Instagram and TikTok, where generic content dies quickly and over-automation becomes visible fast.

What about analytics and memory?

This is where the pairing becomes a real system instead of a content vending machine.

Wonda gives Hermes the execution layer for brand context, research, generation, publishing, and analytics. Hermes gives Wonda a persistent memory and orchestration layer. Together, that means each week's output can shape next week's plan.

Without that loop, the system is just posting.

With it, the system can actually improve:

• keep a stable brand voice
• notice which hooks or formats performed
• stop repeating weak ideas
• build the next batch from the prior week's signal

This is the most underappreciated part of the whole setup. The real leverage is not "the agent can post." The real leverage is "the agent can remember, adapt, and keep the process moving."

Where does Hermes win, and where does Claude Code still win?

Hermes is not a replacement for Claude Code.

Claude Code is better when the work is live, interactive, editor-first, and tightly tied to a development session. Hermes is better when the work should keep running on a server, through messaging, on a schedule, with its own memory and operations loop.

That means Hermes wins when you need:

• always-on execution
• a Telegram or Discord operating surface
• recurring schedules
• a persistent content workflow

Claude Code still wins when you want:

• fast one-off drafting at the keyboard
• interactive exploration in an editor
• a tighter coding workflow around the marketing task

Many teams should use both. But the dividing line should stay clear: if you already live in Claude Code and do not need an always-on daemon, Hermes may be unnecessary. If you want a content system that keeps running without you, Hermes earns its place quickly.

What usually breaks first?

The common failures are straightforward.

Headless auth

If you install on a VPS with no browser, wonda auth login becomes friction. Use WONDA_API_KEY and move on.

Skill not discovered

If Hermes does not see Wonda, confirm the skill is actually present under ~/.hermes/skills/. Most failures here are install or restart issues, not deep runtime bugs.

Rate limits

If you push too much content too quickly, platform constraints become the real bottleneck. Instagram's Graph API publish limits still apply, and the batching article covers those details in depth: How to Batch Create and Schedule Instagram Reels.

Bad abstraction

This is the subtle one. If you treat Hermes as a thin shell over Wonda and keep forcing low-level commands into every prompt, the workflow gets noisier and less useful. The better abstraction is outcome-first: ask Hermes for the content operation you want, not the exact verb chain you think it should call.

Frequently Asked Questions

What is Hermes Agent in one sentence?

Hermes Agent is Nous Research's self-hosted personal agent runtime: an MIT-licensed daemon with persistent memory, messaging integrations, and a skills system that can load published capabilities like Wonda from skills.sh.

How do I install Wonda into Hermes?

Use the Hermes-native install path: hermes skills install degausai/wonda/skills/wonda-cli. That is the path this article recommends.

Do I need to learn Wonda commands first?

No. That is the point of this setup. Wonda is the execution layer; Hermes is the interface you should primarily talk to. If you want the lower-level Wonda surface later, that belongs in separate CLI-focused articles.

Do I need the Wonda MCP server too?

No. Wonda also ships a public MCP server, which is useful if your Hermes setup already standardizes on MCP servers. But the easiest starting point for Hermes users is still the Wonda skill.

Can Hermes post to social media without custom code?

Yes. Once Wonda is installed as a skill, Hermes can operate the publishing workflow without you building the integration yourself.

Is Hermes free?

Yes. Hermes Agent is MIT-licensed and self-hosted. Your real ongoing cost is whichever model provider and media generation you choose to route through.

Can I run this on a cheap VPS?

Yes. For a posting-focused daemon, Hermes itself is not the expensive part. The real cost driver is the model and generation workload you attach to it.

The Bottom Line

The right way to think about this pairing is simple.

Hermes is the operating surface. Wonda is the marketing capability layer. Once the setup is done, the line between them should matter less and less to the user.

That is why the skill matters. It turns "build a marketing agent" into "install one and start using it." The easiest setup is still:

curl -fsSL https://raw.githubusercontent.com/NousResearch/hermes-agent/main/scripts/install.sh | bash
hermes skills install degausai/wonda/skills/wonda-cli
wonda auth login

After that, the useful behavior is not "memorize Wonda." The useful behavior is "use Hermes as the one place where your content pipeline lives."

If you want the editor-first version, start with You Don't Need to Learn the CLI: Let Claude Code Run Wonda for You. If you want the Instagram constraints article, use How to Batch Create and Schedule Instagram Reels. This post is about the layer above both of those: making the whole system feel like one tool.