Ollama Business Memory

In The Fitting Room Memory

Add, edit, delete, and reorder business memory items sent to Ollama.

Jump To Module

Back to Generator Backend Home

Part 2

Add Memory Item

Current Memory Items

Title

Sort Order

Memory Text

Best setup for Ella’s Alterations
Layer 1: Create an Ella’s Alterations custom Ollama model

Use an Ollama Modelfile for the permanent stuff the AI should always know, like business name, tone, role, and response style. Ollama’s Modelfile supports a SYSTEM message, which is exactly where this belongs.

Example:

mkdir -p ~/ollama-ellas
cd ~/ollama-ellas
nano Modelfile

Paste this:

FROM llama3.1:8b

SYSTEM """
You are the local AI assistant for Ella's Alterations, a professional tailoring and alterations studio in Zephyrhills, Florida.

Business identity:
Ella's Alterations specializes in wedding dress alterations, formalwear, bridesmaid dresses, mother of the bride dresses, prom dresses, suits, tuxedos, jacket fit, hems, garment repairs, and custom tailoring guidance.

Voice:
Write in a casual, clear, confident, helpful style. Keep answers practical. Avoid vague advice. Explain things like a master tailor speaking to a real customer.

Important business style:
Mention that fitting appointments matter because formalwear alterations depend on fabric, structure, seams, beadwork, lining, and body fit. Never guarantee a price without seeing the garment. Encourage appointments for bridal and formalwear questions.

Location:
Ella's Alterations serves Zephyrhills, Wesley Chapel, Dade City, Tampa, Lakeland, Orlando, and surrounding Central Florida areas.

SEO style:
When writing website content, use answer boxes, key takeaways, fast answer panels, quick facts, comparison tables, tailor insight, mini glossary, and clear FAQ sections when requested.

Rules:
Do not invent exact prices, dates, awards, customer names, or policies unless they are provided in the knowledge context.
If unsure, say what information is needed.
"""

Then create the model:

ollama create ellas-assistant -f Modelfile
ollama run ellas-assistant

This gives you a reusable model name:

ollama run ellas-assistant
Layer 2: Use RAG for the real “memory”

This is the big one.

For Ella’s Alterations, the AI should be able to search your actual content, like:

services.txt
pricing-guidelines.txt
appointment-policy.txt
bridal-alterations.txt
rush-policy.txt
seo-style-guide.txt
reviews.txt
business-awards.txt
json-ld-rules.txt
common-customer-questions.txt
blog-template-sections.txt

Do not cram all that into the Modelfile. That becomes messy fast.

Use RAG, which means the system searches your documents first, pulls the most relevant pieces, then sends those pieces to Ollama with the user’s question. Ollama’s embeddings documentation says embeddings are used for semantic search, retrieval, and RAG workflows.

Think of it like this:

Customer question
↓
Search Ella’s Alterations knowledge files
↓
Pull the best matching facts
↓
Send facts plus question to Ollama
↓
Answer based on your business knowledge

That is the closest thing to “memory” without retraining.

Layer 3: Add conversation memory

This is for things like:

The user asked about prom dress hems earlier.
The customer said the wedding is next Saturday.
The dress has heavy beading.
The customer is asking for rush service.

That memory should live in your app, not inside Ollama.

Use a small MariaDB or SQLite table like:

CREATE TABLE ai_memory (
    id INT AUTO_INCREMENT PRIMARY KEY,
    customer_name VARCHAR(255),
    memory_type VARCHAR(100),
    memory_text TEXT,
    created_at TIMESTAMP DEFAULT CURRENT_TIMESTAMP
);

Example records:

customer_name: Sarah
memory_type: garment
memory_text: Customer has a beaded mermaid wedding dress and asked about taking in the hips.

customer_name: Sarah
memory_type: deadline
memory_text: Wedding date is May 30, 2026.

Then your app can include relevant memory before sending the question to Ollama.

What I would use on your setup

Since you already have Ollama, Open WebUI, Qdrant, and MariaDB experience, I would do this:

Simple test version

Use:

Ollama
Open WebUI
Knowledge files uploaded into Open WebUI
Custom system prompt for Ella’s Alterations

This is the fastest way to test whether the idea works.

Better local business version

Use:

Ollama for the model
nomic-embed-text or another embedding model for document search
Qdrant for vector storage
MariaDB for customer memory
PHP or Python API wrapper

Ollama has official support for embedding models, and their own blog shows using embedding models for RAG applications.

What should go where
Information	Best Place	Why
Business name	Modelfile	Always needed
Writing tone	Modelfile	Always needed
Location	Modelfile	Always needed
Services	RAG documents	Easy to update
Pricing guidance	RAG documents	Changes over time
Appointment rules	RAG documents	Changes over time
Customer details	Database memory	Private and customer specific
Blog templates	RAG documents	Too large for system prompt
Reviews	RAG documents	Must be accurate
JSON-LD rules	RAG documents	Reusable SEO knowledge
Previous conversation	App memory	Session specific
Do not fine tune it yet

Fine tuning sounds tempting, but for this use case it is overkill.

Fine tuning is better when you need the model to learn a writing pattern or behavior deeply. But for Ella’s Alterations, most of what you need is business facts, policies, services, wording, and local SEO structure. That belongs in RAG.

Tell it like it is: fine tuning for this would be like sewing a dress shut when you only needed a zipper.

Starter folder structure

I would create this:

mkdir -p /mnt/drive1/ellas-ai/knowledge
mkdir -p /mnt/drive1/ellas-ai/prompts
mkdir -p /mnt/drive1/ellas-ai/memory
mkdir -p /mnt/drive1/ellas-ai/scripts

Then files like:

/mnt/drive1/ellas-ai/knowledge/business-profile.txt
/mnt/drive1/ellas-ai/knowledge/services.txt
/mnt/drive1/ellas-ai/knowledge/bridal-alterations.txt
/mnt/drive1/ellas-ai/knowledge/formalwear-alterations.txt
/mnt/drive1/ellas-ai/knowledge/appointment-policy.txt
/mnt/drive1/ellas-ai/knowledge/rush-service.txt
/mnt/drive1/ellas-ai/knowledge/seo-content-style.txt
/mnt/drive1/ellas-ai/knowledge/blog-section-template.txt
/mnt/drive1/ellas-ai/knowledge/json-ld-rules.txt
/mnt/drive1/ellas-ai/knowledge/common-faq.txt