Chat

Overview

ArcLume includes two AI chat interfaces that let you ask questions about your codebase and get answers grounded in your indexed code.

Global chat

Access the global chat from the Chat item in your sidebar. This is a general-purpose assistant that:

• Answers questions about your codebase using semantic search (RAG)
• Retrieves the most relevant code chunks from your indexed repos
• Suggests creating new maps when the conversation leads to feature ideas or improvements
• Maintains conversation history so you can resume previous threads

Map chat

Each map has its own chat tab that provides an AI assistant scoped to that map's context. The map chat can:

• Answer questions about the map's stories and epics
• Create new stories directly from the conversation
• Update existing story details (title, description, acceptance criteria)
• Delete stories
• Update the map's outline and description

The map chat is useful for iterating on your plan conversationally — describe what you want, and the AI creates or modifies stories accordingly.

How RAG works

When you ask a question, ArcLume:

1. Converts your question into a vector embedding
2. Searches your indexed code chunks by vector similarity
3. Retrieves the top 15 most relevant code snippets
4. Passes them as context to Claude along with your question

This means answers reference real files, functions, and patterns from your codebase — not generic information.