Configuration

QALITA Studio offers great flexibility in configuring AI providers and their parameters.

Configuration Architecture

Studio's LLM configuration is managed through QALITA Platform's Settings > AI Configuration interface. Configurations are stored in the Platform database and linked to your organization (partner).

Configuration Model

Each LLM configuration includes:

Field	Description
name	Display name for the configuration
provider	Provider type (openai, anthropic, ollama, etc.)
model_name	Specific model to use
api_key	API key for authentication (encrypted)
endpoint_url	Custom endpoint URL (for Ollama, Azure, generic)
is_active	Whether this is the active configuration
configuration	Additional parameters (temperature, timeout, etc.)

Multiple Configurations

You can create multiple LLM configurations for different use cases:

• Development: Use Ollama for free local testing
• Production: Use GPT-4o for high-quality responses
• Cost-optimized: Use GPT-4o-mini for routine tasks

Available Providers

1. Ollama (Local)

The local provider uses Ollama to run open-source models locally.

Advantages:

• ✅ Free and unlimited
• ✅ Total privacy (local data)
• ✅ No network dependency
• ✅ Many models available

Platform Configuration:

Field	Value
Provider	Ollama
Endpoint URL	http://localhost:11434 (or your host)
Model	llama3.2, qwen2.5:7b, etc.
API Key	Not required

Recommended Models:

Model	Size	Usage	Performance
qwen2.5:7b	7B	General, multilingual	⭐⭐⭐⭐⭐
llama3.2	3B	Fast, lightweight	⭐⭐⭐⭐
mistral:7b	7B	Reasoning	⭐⭐⭐⭐⭐
phi3:medium	14B	Precise	⭐⭐⭐⭐
deepseek-coder:6.7b	6.7B	SQL/Python code	⭐⭐⭐⭐⭐

Installing a model:

# Download a model
ollama pull qwen2.5:7b

# List installed models
ollama list

# Test a model
ollama run qwen2.5:7b "Hello!"

Verification:

Test endpoint: http://127.0.0.1:11434/api/tags

Network Access

Ensure the Platform backend can reach the Ollama server. If Ollama runs on a different host, use its IP address or hostname.

2. OpenAI (ChatGPT)

Access GPT-4, GPT-4o, and GPT-3.5 models.

Platform Configuration:

Field	Value
Provider	OpenAI
API Key	sk-proj-...
Model	gpt-4o, gpt-4o-mini, gpt-3.5-turbo
Endpoint URL	Not required (uses default)

Available Models:

Model	Context	Cost	Usage
gpt-4o	128K	$$	Best quality/price ratio
gpt-4o-mini	128K	$	Fast and economical
gpt-3.5-turbo	16K	$	Basic, fast

Getting an API key:

1. Create an account on platform.openai.com
2. Go to API Keys
3. Create a new secret key
4. Copy it (it won't be visible again)

Pricing: Check openai.com/pricing

3. Azure OpenAI

Enterprise-grade OpenAI models on Azure infrastructure.

Platform Configuration:

Field	Value
Provider	Azure OpenAI
Endpoint URL	https://your-resource.openai.azure.com/
API Key	Your Azure OpenAI API key
Model	Your deployed model name

Advantages:

• Enterprise compliance and security
• Regional data residency
• SLA guarantees
• Integration with Azure services

Getting started:

1. Create an Azure OpenAI resource in Azure Portal
2. Deploy a model (gpt-4o, gpt-4, etc.)
3. Get the endpoint URL and API key from the resource

4. Mistral AI

High-quality French models.

Platform Configuration:

Field	Value
Provider	Mistral
API Key	From console.mistral.ai
Model	mistral-large-latest, mistral-small-latest

Available Models:

Model	Context	Performance
mistral-large-latest	128K	Excellent
mistral-small-latest	32K	Good
open-mistral-7b	32K	Decent

Getting an API key:

1. Sign up on console.mistral.ai
2. Create an API Key
3. Top up your credits if necessary

Advantages:

• Excellent for French language
• Good context understanding
• Competitive pricing

5. Claude (Anthropic)

Claude 3 models for advanced reasoning.

Platform Configuration:

Field	Value
Provider	Anthropic
API Key	sk-ant-... from console.anthropic.com
Model	claude-3-5-sonnet-20241022, claude-3-opus-20240229

Available Models:

Model	Context	Capabilities
claude-3-5-sonnet-20241022	200K	Versatile excellence
claude-3-opus-20240229	200K	Complex reasoning
claude-3-haiku-20240307	200K	Fast and lightweight

Getting an API key:

1. Create an account on console.anthropic.com
2. Go to API Keys
3. Generate a new key

Features:

• Excellent for complex analysis
• Very good in multiple languages
• Context up to 200K tokens

6. Generic Provider (OpenAI-compatible) 🔧

Connect to any OpenAI-compatible API endpoint (vLLM, LM Studio, etc.).

Platform Configuration:

Field	Value
Provider	Generic
Endpoint URL	Your server URL (e.g., http://localhost:8000/v1)
API Key	Server API key (if required)
Model	Model name as expected by the server

Use cases:

• Self-hosted LLM servers (vLLM, text-generation-inference)
• LM Studio local deployment
• Custom fine-tuned models
• Air-gapped environments

Configuration via Platform Interface

Add a Configuration

1. Go to Settings in Platform
2. Navigate to AI Configuration
3. Click Add Configuration
4. Fill in the required fields:
   • Name: Display name for this configuration
   • Provider: Select from dropdown
   • API Key: For cloud providers
   • Model: Exact model name
   • Endpoint URL: For Ollama, Azure, or Generic
5. Click Test Connection to verify
6. If ✅ success, click Save

Activate a Configuration

1. In AI Configuration list
2. Click on the desired configuration
3. Toggle Active to enable
4. Only one configuration can be active at a time

Edit or Delete

1. Click on a configuration to edit
2. Use the Delete button to remove

Configuration via API

LLM configurations are managed through the Platform's standard REST API.

Get Agent Capabilities

GET /api/v1/studio/agent/capabilities

Returns the current agent status and active configuration:

{
  "agent_available": true,
  "llm_configured": true,
  "active_config": {
    "id": 1,
    "name": "Production OpenAI",
    "provider": "openai",
    "model_name": "gpt-4o-mini",
    "endpoint_url": null
  }
}

Get Studio Status

GET /api/v1/studio/status

Returns worker connectivity status:

{
  "connected_workers": [1, 3],
  "worker_count": 2
}

Backend Dependencies

For the agent module to be available, install the required Python packages on the Platform backend:

pip install langchain langgraph langchain-openai langchain-anthropic langchain-ollama

These are optional dependencies. If not installed, the agent_available field will be false.

Advanced Configuration

Custom Models (Ollama)

You can create your own models with Ollama:

# Create a Modelfile
cat > Modelfile <<EOF
FROM qwen2.5:7b
SYSTEM "You are a data quality expert. Always answer professionally."
PARAMETER temperature 0.7
PARAMETER top_p 0.9
EOF

# Create the model
ollama create qalita-expert -f Modelfile

# Use in Platform configuration
# model: "qalita-expert"

Additional Configuration Parameters

The configuration field in Platform supports extra parameters:

{
  "temperature": 0.0,
  "timeout_seconds": 60.0,
  "max_retries": 2
}

Parameter	Type	Default	Description
temperature	float	0.0	Creativity (0.0 = deterministic, 1.0 = creative)
timeout_seconds	float	60.0	Request timeout
max_retries	int	2	Retry attempts on failure

Security

API Key Protection

• API keys are stored encrypted in the Platform database
• Keys are never exposed in API responses
• Access is controlled by Platform authentication

Data Privacy

Studio conversations may include:

• Data samples from sources
• Quality metrics and recommendations
• Schema information

Ensure your LLM provider's data handling policies align with your organization's requirements. For maximum privacy, use Ollama with local models.

Troubleshooting

"Agent module not available"

Install the required dependencies on the Platform backend:

pip install langchain langgraph langchain-openai langchain-anthropic langchain-ollama

"No LLM configuration found"

1. Go to Settings > AI Configuration
2. Create a new configuration
3. Set it as Active

"Model not found in Ollama"

# Check that the model is installed
ollama list

# Install it if necessary
ollama pull <model-name>

Authentication error (401/403)

• Verify that your API key is valid
• Test directly with the provider's API
• Check that you have available credits

Connection timeout

Check network connectivity from the Platform backend to the LLM provider:

# For Ollama
curl http://localhost:11434/api/tags

# For OpenAI
curl https://api.openai.com/v1/models \
  -H "Authorization: Bearer YOUR_KEY"

Best Practices

1. Start with Local: Test with Ollama first before using paid APIs
2. Adapted Models: Choose the model according to your use case
3. Cost Monitoring: Use gpt-4o-mini for development and testing
4. Security: Rotate API keys periodically
5. Multiple Configs: Create configurations for different environments

Next Steps

• 🚀 Features - Explore all Studio capabilities
• 💬 Conversation Management - Organize your conversations
• 🔧 Platform Integration - Deep dive into Platform integration