Sessions

Sessions capture the sequences of events in your AI application for a given user interaction or task completion. For example, a RAG pipeline can be tracked from the initial query, to the retrieval step, reranker, and final Prompt generation. Or a chatbot agent’s interaction with a user can be tracked through the whole conversation and tool calls.

Understanding Sessions

A Session is essentially a sequence of Logs. Each Log has inputs, outputs, and metadata. The Session is as simple as having a common session_id for all the Logs in a Session and an optional parent-child relationship between Logs.

These Logs can represent various types of interactions and operations within your AI application, including:

• Prompt executions
• Tool calls
• Evaluator runs
• Other custom events

Sessions are particularly useful for tracking the behavior of AI agents or complex chains of operations, as they allow you to see the entire context and flow of a particular interaction or task.

Use Cases for Sessions

Sessions are particularly valuable in the following scenarios:

1. Agent Workflows: When working with AI agents that make multiple decisions or take multiple actions, Sessions help you trace the agent’s thought process and actions.

2. Complex Chains: For applications that involve chains of operations (e.g., a series of Prompt calls with intermediate processing), Sessions provide a clear view of the entire process.

3. Debugging and Optimization: By examining Session logs, you can identify bottlenecks, errors, or unexpected behaviors in your AI workflows.

4. Performance Analysis: Sessions allow you to analyze the efficiency and effectiveness of your AI processes over time.

Working with Sessions

To leverage Sessions in your application, you can follow our guide on chaining calls and sessions. The process can be as simple as setting a session_id in your Log or Call API requests. Here’s a basic overview:

1. Generate a unique session_id at the beginning of a workflow or user interaction.
2. Include this session_id when logging each significant step or operation within the Session.
3. For complex workflows, you can create nested Sessions by specifying parent-child relationships between logs.
4. Continue using the same session_id until the workflow or interaction is complete.

By consistently using Sessions in this manner, you’ll gain valuable insights into your system’s behavior, making it easier to debug, optimize, and improve your AI-driven processes. Refer to the guide for detailed implementation steps and best practices.