On-brand agents with Hermes Desktop.

AI agents are getting good at doing work. Writing copy, scheduling posts, drafting campaigns. What they're not good at is looking like your brand while they do it.

The problem isn't the model. It's that the brand lives in a PDF, a Figma file, and someone's head. So the agent fills the gap with an average of the internet, and the output drifts off brand every time.

Bloom fixes that. Bloom is the brand layer, the canonical home of your brand's identity, callable by any system that creates on its behalf. When an agent needs an asset, it calls Bloom, and what comes back is on-brand.

This guide walks through connecting Bloom to Hermes Desktop, the native desktop app from Nous Research for running AI agents locally. By the end you'll have an agent that can generate on-brand assets from a single prompt.

What is Hermes Desktop?

Hermes is an open-source AI agent by Nous Research. It comes in three flavors: terminal, web dashboard, and desktop app. All share the same settings and sessions.

Hermes Desktop is the app version. It handles model selection, server management, and tool approval through a visual interface. You pick a model, connect your tools, and start chatting.

What makes it useful for Bloom: Hermes supports MCP natively. Connecting Bloom gives the agent direct access to your brand. It can onboard brands, generate on-brand assets, edit them, resize for different platforms, and search your asset library. All from a single conversation.

What you'll need

• Hermes Desktop. Download from hermes-agent.nousresearch.com
• A Bloom account. Sign up at trybloom.ai if you don't have one
• A model API key from Anthropic, OpenAI, or any provider Hermes supports (Hermes also supports Nous Portal with 300+ models)

Setup

Four steps. The whole thing takes about five minutes.

1

Install Hermes Desktop

Download the installer for your platform from the Hermes site. On first launch, Hermes installs everything it needs on your computer and walks you through picking a model provider.

2

Pick a model

The first-launch wizard asks you to choose a provider and model. Any model with 64K+ context works. We recommend Claude Sonnet or Opus for the best tool-use performance with Bloom.

You can change models any time from the picker in the prompt box.

hermes mcp add bloom --url "https://www.trybloom.ai/api/mcp" --auth oauth

4

Verify the connection

Ask Hermes if Bloom is working:

List all Bloom MCP tools.

You should see bloom_generate_image, bloom_onboard_brand, bloom_list_brands, and others. If they show up, you're connected.

Using it

Once Bloom is connected, the agent can call your brand directly. Here are a few things you can do from a single message.

Onboard a brand

If you haven't set up your brand in Bloom yet, the agent can do it for you:

Onboard gumroad.com into Bloom.

Hermes calls bloom_onboard_brand, and Bloom pulls in your logo, palette, typography, and visual style. A minute later the brand lives in Bloom, callable by the agent.

Generate on-brand assets

Generate three Instagram post assets for Gumroad's creator tools launch.

The agent calls Bloom with your brand context. Every asset comes back matching your palette, typography, and visual style. No design brief needed.

Recreate a high-performing ad

Find high-performing Instagram ads for creator platforms, then recreate the best one for Gumroad.

Bloom keeps a curated library of real, high-performing ads. The agent calls bloom_find_reference_ads to surface relevant examples, then recreates them on-brand. A proven format with your identity, not a blank-canvas guess.

Edit and resize

Resize the best one for X and LinkedIn too, keep the Gumroad brand.

Bloom handles resizing and format adaptation. The agent can chain generate, edit, and resize in a single conversation without you opening another tool.

What the agent gets access to

When you connect Bloom over MCP, Hermes can call any of these tools in conversation:

• bloom_generate_image: generate on-brand assets from a text prompt
• bloom_edit_image: edit an existing asset with brand-aware instructions
• bloom_resize_image: adapt images for different platforms and aspect ratios
• bloom_find_reference_ads: surface real, high-performing ads to recreate on-brand
• bloom_onboard_brand: create a new brand from a website URL
• bloom_list_brands: see all brands in the account
• bloom_search_user_images: find existing assets to use as references
• bloom_remove_background: clean background removal
• bloom_vectorize_image: convert raster assets to SVG

The full list is in the Bloom MCP docs.

Going further

This setup gives you a local agent that can generate on-brand content on demand. A few ways to extend it:

• Connect more tools. Hermes supports any MCP-compatible tool. Add Slack, Linear, Notion, or your own tools alongside Bloom and the agent can publish, file tickets, or update docs with on-brand assets in one pass.
• Use reference assets. Call bloom_search_user_images before generating to pull existing assets as references. This consistently improves output quality.
• Run it remote. Hermes Desktop can connect to a remote Hermes instance running on a VPS. Run the heavy compute server-side and use the desktop app as a thin client.

Bloom, the brand layer, callable from any agent. trybloom.ai

Hermes Desktop, open-source AI agent with native MCP support. hermes-agent.nousresearch.com