Session Management

Sessions are the foundation of Shadai. They organize your documents, maintain conversation context, and enable powerful multi-document workflows.

What is a Session?

A session is a container that holds:

📚 Documents - Your ingested files

💬 Chat history - All questions and answers

🧠 Context - Conversation memory

⚙️ Configuration - Session-specific settings

Think of a session as a workspace for a specific project or topic.

Session Types

Named Sessions (Persistent)

Perfect for ongoing projects:

Use named sessions when:

Working on ongoing projects

Building up knowledge over time

Need to preserve history

Collaborating with team members

Temporal Sessions (Auto-Delete)

Perfect for one-off queries:

Use temporal sessions when:

One-time queries

Testing or experiments

Privacy-sensitive data

No need to persist data

Creating Sessions

Basic Session Creation

Session Naming Best Practices

✅ Good Names:

❌ Bad Names:

Model Configuration

You can configure which LLM and embedding models your session uses. This allows you to choose the best model for your specific use case based on performance, cost, and capabilities.

Specifying Models

Available Models

LLM Models (31 options):

OpenAI: GPT-5, GPT-4.1, GPT-4o series

Azure: Same models via Azure deployment

Anthropic: Claude Sonnet 4.5, Claude Opus 4, Claude Haiku 3.5

Google: Gemini 2.5 Pro, Gemini 2.0 Flash series

Embedding Models (5 options):

OpenAI: text-embedding-3-large, text-embedding-3-small

Azure: Same embeddings via Azure

Google: gemini-embedding-001

See complete model list →

Model Selection Examples

Cost-Effective Setup

Premium Performance

Creative Tasks

Mixed Providers

Enterprise (Azure)

System Prompts

Customize your session's behavior with system prompts:

Model Selection Best Practices

Choose your LLM based on:

Task complexity: GPT-4o/Claude Opus for complex reasoning, Gemini Flash for simple tasks

Speed requirements: Gemini Flash models are fastest

Cost constraints: Google models generally most cost-effective

Specific capabilities: Claude for creative work, GPT-4 for technical analysis

Choose your embedding model based on:

Quality needs: text-embedding-3-large for best retrieval accuracy

Speed/cost: text-embedding-3-small or Gemini embeddings for efficiency

Provider alignment: Match with your LLM provider when possible

System prompt tips:

Be specific about role and expertise

Include output format preferences

Mention any constraints or guidelines

Keep it concise but comprehensive

Session Lifecycle

1. Session Creation

What happens:

🔍 Checks if session exists

✨ Creates new if doesn't exist

📂 Loads existing data if exists

2. Adding Documents

Documents are cumulative - new ingestions add to existing documents.

3. Querying

Each query:

Searches ALL documents in session

Maintains conversation history

Updates context

4. Session Cleanup

Working with Multiple Sessions

Pattern 1: Sequential Processing

Pattern 2: Parallel Processing

Pattern 3: Hierarchical Sessions