Agent Chat

Overview

The Agent Chat is Keystone's primary interface for querying your codebase knowledge. It uses retrieval-augmented generation (RAG) to provide answers grounded in your actual repository history.

Using the chat

Open the Agent Chat from your project page. Type your question in natural language and Keystone will:

1. Search its knowledge base for relevant context
2. Generate a response based on actual commits, diffs, and decisions
3. Provide references to the source material

Types of questions

Decision queries

Ask about the reasoning behind specific decisions:

• "Why did we choose PostgreSQL over MongoDB?"
• "What was the motivation for the recent auth refactor?"

Architecture queries

Understand how your system is structured:

• "How does the data pipeline work?"
• "What's the relationship between the user and organization models?"

History queries

Trace the evolution of specific parts of your codebase:

• "How has the API authentication changed over time?"
• "What recent changes were made to the payment module?"

Pattern queries

Discover conventions and patterns:

• "What testing patterns do we use?"
• "How do we handle error responses in the API?"

Tips for better results

• Be specific: "Why did we add rate limiting to the /api/search endpoint?" is better than "Tell me about rate limiting"
• Reference components: mention specific files, modules, or features when possible
• Ask follow-ups: the chat maintains context, so you can dig deeper into any answer
• Ask about trade-offs: Keystone excels at explaining what alternatives were considered