Understanding Models, Creativity Levels, and Embeddings—The nemo Way
- nemo
- Aug 26
- 3 min read
Updated: Aug 27
You don’t need to be an AI expert; just put on your strategic hat.
Creating an AI Agent with nemo begins with strategy. After all, the most important thing isn't the technology, but the people operating it. With strategy in mind, you start by giving your agent a job, then defining its personality.
What's next?
Shape How Your Agent Thinks
Depending on your agent's "job description" and how you want it to interact with the people it serves, you can decide how creative your agent should be, how it understands your content, and which AI “brain” it uses.
This means choosing:
The generative model (aka the LLM)
The embedding model
The creativity level
Let's look at the basics of setting up your AI agent (these steps are part of the first section of the nemo AI Agent Building App).
What Is a “Generative Model (or LLM)”?
One of the first tasks in the nemo agent-creation process is selecting an LLM model and version. This is like selecting a “brain” for your agent, and it influences the text replies when someone asks a question.
In nemo, you’ll see several model options. Each one is a slightly different version of a large language model (LLM), like those created by OpenAI (ChatGPT), Google (Gemini), Anthropic (Claude), and others.
How to Choose a Generative Model:
If you’re not sure, ask Aura.
If your agent needs faster answers or you want to test cost differences later, Aura (your AI sidekick) can help you compare.
If accuracy and safety are top priorities (like in HR or compliance), ask Aura for the most stable model.
Tip from nemo: Don’t overthink it—your training content and instructions will influence behavior more than the model you choose.
For a deeper dive into generative models, check out our article on selecting the best AI brain for your Agent
What’s an “Embedding Model”?
This model helps your agent find the right content in your training data.
When someone asks a question, your agent looks at the training documents you uploaded. The embedding model helps it understand:
Which parts of the content are most relevant
How to connect a user’s question with the right answer
Think of it like a librarian who doesn’t just search for exact words, but understands the meaning of what you’re asking.
How to Choose an Embedding Model:
Consider how many documents your agent will have to reference, and how complex they are
If your agent will reference a lot of long documents, you can test a more advanced embedding model for better search quality. If the reference documents are fairly straightforward, you may be ok with a low-token, less advanced model.
Ask Aura -- she's there to help!
And for more details, read our article about how AI looks up information (Retrieval Augmented Generation)
What’s the “Creativity Level”?
This setting adjusts the balance between accuracy and creativity in your agent’s responses.
In nemo, it’s a simple slider:
Low creativity (10-30) = more accurate, to-the-point answers
High creativity (80-100) = more expressive, conversational answers (with slightly higher risk of “guessing”)
Everything in between (40-70) = the middle of the road (with a balanced blend of expression and accuracy)
When to Go Lower:
HR, compliance, support agents
Agents that require strict accuracy
FAQs or structured knowledge
When to Go Higher:
Marketing, sales, or engagement agents
Agents meant to sound lively, witty, or branded
Creative writing, ideation, or brainstorming bots
Examples:
Low creativity: “The return period is 30 days.”
High creativity: “No worries—our return policy has you covered!”
What If I Pick the “Wrong” Settings?
You won’t break anything. You can always:
Ask Aura for help
Change the settings later
Test different setups using preview mode
Don’t be afraid to experiment. Change one variable, test out the Agent’s answers, and adjust if needed.
nemo is designed to let you experiment safely.
What’s Next?
Now that your agent has a voice, knowledge, and brainpower—it’s time to preview how it works and get ready to share it with the world.
Comments