Create Session Summarization API

To enhance long-term memory recall and provide high-level overviews of past work, this ticket covers the creation of a script that can summarize a completed agent session.

This script will serve as a data processing tool that can be run independently, without needing a full interactive agent session. It will read all the memories for a given session, use a generative model to create a concise summary, and store that summary in a new sessions database collection.

This creates a powerful, two-tiered memory system: a high-level index of session summaries, and a detailed log of individual interactions.

Workflow in Practice

1. Initiation: The process is triggered via a command like npm run ai:summarize-session -- --session-id <ID_of_session>.
2. Memory Retrieval: The script connects to the neo-agent-memory collection and retrieves all documents for the given session ID.
3. Content Aggregation: It concatenates the prompt, thought, and response from all memories into a single text block.
4. Summarization: It sends this text to the Gemini API with a prompt to summarize the key goals, decisions, and outcomes of the session.
5. Persistence: The resulting summary is saved as a new document in a new sessions collection in ChromaDB.

Acceptance Criteria

1. A new script is created (e.g., buildScripts/ai/summarizeSession.mjs).
2. The script accepts a sessionId as a command-line argument.
3. A new ChromaDB collection (e.g., neo-agent-sessions) is created to store the summaries.
4. The script successfully retrieves memories, generates a summary via the Gemini API, and stores it in the new collection.
5. The aiConfig.mjs file is updated with configuration for the new sessions collection.