Chat Interface

The Chat page lets you ask questions about your indexed repositories in plain English. Superpositional searches your codebase — code structure, relationships, and content — then uses LLM reasoning to produce an answer with references to the specific files and functions it found.

How to use it

1. Open the Chat page from the sidebar.
2. Type a question in the input field.
3. Superpositional streams the response in real time. You'll see tool calls as they happen (searches, entity lookups, dependency traversals) followed by the synthesised answer.

Chat sessions are persistent. You can return to previous conversations from the session list in the sidebar. You can also start a new session or delete old ones.

What the chat can do

The chat agent has access to several tools that it uses automatically based on your question:

• Code search. Semantic search across your indexed repositories. Finds relevant code by meaning, not just keywords.
• Entity lookup. Look up specific functions, classes, modules, and files by name.
• Dependency traversal. Explore the relationship graph — what calls what, what imports what, what depends on what.
• PR change inspection. Examine the file changes in a specific pull request.
• Document and claims search. Search across generated documents and the structured claims extracted from them.
• Issue listing. List open issues tracked by Superpositional.
• Document generation. Request Superpositional to generate a document (summary, changelog, architecture doc) for a repository.
• Deep analysis. For complex questions, the chat can escalate to a full reasoning pipeline that does multi-step investigation with planning, evidence gathering, and synthesis.

Example questions

Here are some questions that work well:

Architecture and structure:

• "How is authentication implemented in this project?"
• "What are the main entry points for the API?"
• "Which modules depend on the database layer?"

Dependencies and impact:

• "What does the payments service depend on?"
• "Which files import the Logger class?"
• "What would break if I changed the signature of processOrder?"

Code search:

• "Where is the retry logic for failed queue messages?"
• "Find all error handling in the webhook handler"
• "How is user input validated before database writes?"

PR changes:

• "What files changed in PR #42?"
• "Show me the changes in the latest pull request for this repo"

The more specific your question, the more useful the answer. Include file names, function names, or module names when you know them — it helps the agent narrow the search.

Tips

• Ask follow-ups. The chat keeps context within a session. If the first answer is close but not quite right, ask a follow-up to refine it.
• Scope to a repo. If you have multiple repositories indexed, mention which one you're asking about. Otherwise the agent searches across all of them.
• Watch the tool calls. The chat shows which tools the agent is using (search, entity lookup, dependency traversal, etc.) as it works. This gives you visibility into how it's arriving at the answer.

Limitations

• The chat reasons over your indexed codebase. It does not execute code or access running services.
• Answers are based on the current index. If your repository hasn't been re-indexed since a recent push, the chat may not reflect the latest changes.
• The chat agent runs for up to 10 tool-use iterations per question. Very broad questions may hit this limit before producing a complete answer. If that happens, ask a more focused follow-up.