Skip to main content
Answer customer questions by asking Agent to investigate your company’s codebase. Get plain-language explanations of product behavior, backed by code references, so you can prepare a clear reply or a focused escalation to engineering.

Best for

  • Support agents answering questions that the help center does not cover.
  • Support leads preparing a focused engineering escalation.

Connectors

Before you start

You need a Replit Pro or Enterprise plan and access to your company’s GitHub repository. Connect GitHub with access to the relevant repository. Gather the customer’s expected result, actual behavior, reproduction steps, and error message.

Prompt the agent

What to expect

Agent uses the customer’s question or report to find the relevant parts of the codebase and investigate how the product behaves. If details are missing, Agent asks for clarification or suggests questions to ask the customer. The response includes:
  • A plain-language explanation: how the relevant feature works and what could explain the reported behavior.
  • Supporting references: links to the files and functions behind the findings, so you or an engineer can verify them.
  • Checks to narrow down the cause: information or reproduction steps needed to distinguish between possible explanations.
  • A draft customer reply: a concise explanation and next steps based on the findings, ready for you to review before sending.
You can continue in the same chat with additional customer details, ask Agent to investigate another possibility, or refine the reply. Agent uses the conversation’s context and checks more of the code when needed. If the evidence does not establish the cause, Agent explains what remains unknown. You can ask for an engineering handoff with reproduction steps, expected and actual behavior, relevant code references, and the checks still needed.

Make it work for you

Troubleshooting

Ask Agent to rank the next checks by how easily they distinguish between possible causes.
Confirm the deployed version and feature settings with engineering.
State what remains unknown and request the specific logs or account checks through approved channels.

Turn feedback into improvements

Take a confirmed problem from investigation to a proposed change.

Push a PR from Slack

Request a focused change from the conversation where your team works.