Training Overview
The Training section lets you choose how your AI agent is prepared to understand your product, your data, and your users. Navigable AI supports multiple training modes — from quick setup to deeply customized fine-tuning — so you can balance cost, control, and accuracy.
⚙️ Pick the setup that fits your use case
Each mode represents a different level of customization and training depth. You can switch between modes in the Training tab view of the app.
| Mode | Description | Best For |
|---|---|---|
| Fine-tuning | Trains a new model specifically on your product data. Can optionally use RAG for grounding in your knowledge base. | Deep customization and highest accuracy when RAG is enabled |
| RAG + Standard Model | Uses your knowledge base to ground responses without training the model itself. | Accurate, dynamic responses with low cost |
| Simple (Foundational) | Uses the base model directly — no training or RAG required. | Quick deployment or early prototyping |
🧩 Fine-tuning
Fine-tuning is the most advanced setup for customizing your assistant. It trains a new model using your Q&A dataset, giving your agent a deep understanding of your product, terminology, and behavior.
If RAG is also enabled, your fine-tuned model can dynamically ground its responses in your knowledge base — combining precision and freshness.
Highlights:
- Highest accuracy and reliability
- Learns from your curated product data
- Optional RAG for hybrid grounding
- Trackable training progress and cost
📘 Learn more: Fine-Tuning Form Guide →
📚 RAG (Retrieval-Augmented Generation)
RAG agents don’t require fine-tuning — instead, they use retrieval to pull relevant information from your indexed Q&A knowledge base at runtime. This lets your agent stay current with evolving documentation, policies, or FAQs.
To create a RAG agent, enable:
“Enable indexing for RAG grounded responses”
This indexing process semantically structures your Q&A data, improving recall, accuracy, and grounding.
Highlights:
- Works instantly with your knowledge base
- Reduces hallucinations by referencing real content
- Keeps responses current without retraining
- Lowest total cost among grounded options
📘 Learn more: RAG Form Guide →
⚡ Simple (Foundational Model)
Simple agents use the base model directly without any training or RAG indexing. They’re ideal for quick testing, internal assistants, or early prototypes.
You can still shape their behavior using Custom Prompts — guiding tone, structure, or context.
Highlights:
- Fastest setup — no training required
- Uses the general knowledge of the foundational model
- Fully functional and customizable via prompts
📘 Learn more: Simple Form Guide →
🧭 Choosing the Right Setup
Use this quick reference to decide which setup suits your goals:
| Goal | Recommended Setup | Notes |
|---|---|---|
| You want the most accurate, domain-specific model | Fine-tuning + RAG Enabled | Enable RAG for hybrid grounding & keeping content fresh |
| You need grounded, always-updated answers | RAG + Standard Model | Retrieval keeps content fresh |
| You want a fast, low-cost assistant | Simple | Great for testing or demos |
Check out our pricing estimator to choose the most cost-effective option for your needs: Pricing Estimator →
🪄 Pro Tips
- You can enable RAG at any stage — even after fine-tuning — to combine training and retrieval power.
- Evaluate before deployment — Navigable’s built-in Evaluation Suite helps ensure your model meets your quality standards.
- Cost efficiency: RAG-only is typically 2× runtime cost of a base model, while fine-tuning involves a one-time training cost.
- Add Custom Prompts later to tailor style, tone and response behavior.
🚀 Next Steps
Once you’ve chosen your training setup:
- Complete the relevant form under your selected tab (Fine-tuning, RAG, or Simple).
- Start training (if applicable).
- Track progress in the Training Timeline.
- Move to the Evaluation section to test performance before deploying.