Getting Started from Scratch

Build your own AI assistant with Ada.

This guide walks you through customizing Ada to create YOUR personalized AI assistant - not just use the default.

Overview

Ada ships with a default “helpful assistant” personality named Ada. But the whole point is to make it YOURS:

  • Different name and personality

  • Custom knowledge and context

  • Tailored tone and behavior

  • Your choice of model

Time required: 15-30 minutes Difficulty: Beginner-friendly

Step 1: Get Ada Running

If you haven’t already:

# Install Ollama
curl -fsSL https://ollama.com/install.sh | sh
ollama serve

# Clone and setup
git clone https://github.com/luna-system/ada.git
cd ada
python3 ada_main.py setup

# Pull a model and start
ollama pull qwen2.5-coder:7b
ada run

Checkpoint: Visit http://localhost:7000/v1/healthz - you should see status “ok”.

Step 2: Choose Your AI Model

Ada uses Ollama to run local models. You can use ANY model Ollama supports.

See Available Models

Visit https://ollama.com/library for the full list.

Popular choices:

  • llama3.1 - Meta’s open model, good balance of speed/quality

  • mistral - Fast and capable, great for coding

  • qwen2.5 - Excellent multilingual support

  • qwen2.5-coder - Great coding-focused model

  • deepseek-r1 - Optional, shows reasoning process (can be slower)

  • gemma2 - Google’s open model, efficient

Change the Model

Edit .env in the project root:

# Change this line
OLLAMA_MODEL=llama3.1

# Or use a specific size
OLLAMA_MODEL=mistral:7b

Restart Ada:

ada stop
ada run

The first message will be slower as Ollama downloads the new model.

Checkpoint: Chat with your chosen model and verify it responds.

Step 3: Give Your AI a Personality

Choose an Example or Start Fresh

Ada includes example personas in examples/personas/:

# See what's available
ls examples/personas/

# Copy one you like
cp examples/personas/coding-buddy.md persona.md

# Or start minimal
cp examples/personas/minimal.md persona.md

Edit the Persona

Open persona.md in your editor and customize:

# My AI Assistant

## Role

You are [NAME], a [DESCRIPTION] for [USER].

## Tone

[How should your AI communicate? Formal? Casual? Technical?]

## Priorities

What matters most in responses:
- Be concise / Be detailed
- Focus on code / Focus on ideas
- Ask clarifying questions / Make reasonable assumptions

## Special Context

[Any specific knowledge your AI should have?]
- Your work domain
- Your preferences
- Things to remember about you

Tips:

  • Keep it under 2000 characters (longer personas get truncated)

  • Be specific - “prefer functional programming” works better than “likes code”

  • Use examples to show tone rather than describing it

  • Don’t worry about perfection - iterate!

Reload the Persona

docker compose restart brain

Checkpoint: Chat with your AI. Does it match the personality you defined?

Step 4: Name Your AI

Edit .env:

AI_NAME=Jarvis
AI_USER_NAME=Tony

Restart:

docker compose restart brain

Your AI will now introduce itself as Jarvis and recognize you as Tony.

Checkpoint: Ask “What’s your name?” and verify it responds correctly.

Step 5: Tweak RAG Settings (Optional)

RAG (Retrieval-Augmented Generation) controls how much context/memory your AI uses.

Common Adjustments

In .env:

# How many past conversation turns to include
RAG_TURN_TOP_K=4  # Default: 4, try 6-8 for more context

# How many memories to retrieve
RAG_MEMORY_TOP_K=3  # Default: 3

# Max persona length
RAG_PERSONA_MAX_CHARS=2000  # Default: 2000

# Disable features you don't need
RAG_ENABLE_FAQ=false  # If you don't have FAQs loaded
RAG_ENABLE_SUMMARY=false  # If you don't want auto-summaries

Restart brain after changes.

When to Adjust

Increase context if:

  • Your AI forgets important details from earlier in the conversation

  • You have long, complex discussions

Decrease context if:

  • Responses are slow

  • Your AI seems confused by too much information

  • You want more focused, concise responses

Step 6: Test Core Features

Memory

Have a conversation, then restart the service:

docker compose restart brain

In a new conversation, reference something from before. Your AI should remember (memories are stored in ChromaDB).

OCR (Optional)

Upload an image with text via the web interface. Your AI should extract and discuss the text.

Common Issues

“Model not found”

Ollama hasn’t downloaded the model yet. Wait for the brain container logs:

docker compose logs brain -f

You’ll see download progress.

Persona not loading

Check that persona.md exists and is less than RAG_PERSONA_MAX_CHARS.

View brain logs for errors:

docker compose logs brain | grep persona

Responses are slow

  • Try a smaller model (mistral:7b vs mistral:70b)

  • Reduce RAG_TURN_TOP_K and RAG_MEMORY_TOP_K

  • Check if you have a GPU being used (run docker compose logs ollama)

AI doesn’t remember anything

Check ChromaDB is running:

docker compose ps chroma

If stopped, start it:

docker compose up -d chroma

Next Steps

Now that you have a personalized AI:

  1. Build a specialist - Add custom capabilities

  2. Explore the API - Integrate with other tools

  3. Tune performance - Optimize for your hardware

  4. Share your persona - Help others learn

    • Submit a PR with your persona to examples/personas/

Need Help?

Remember: Ada is a starting point, not a finished product. Your weird ideas are the point! 🚀