Troubleshoot AI in Visual Studio Code

When chat, agents, or inline suggestions do not behave as expected, start with the symptom you see. This article helps you choose the right diagnostic tool, interpret the result, and collect useful evidence if you need to report the issue.

Choose a troubleshooting path

Start with basic checks

Before you collect detailed logs:

1. Run Code: Check for Updates from the Command Palette to update VS Code.
2. Confirm your access method. Sign in with the GitHub account that has your Copilot plan, or select a configured bring-your-own-key model for supported chat features. Some features require the GitHub Copilot service. For details, see GitHub Copilot subscription questions.
3. Open the Copilot status dashboard in the Status Bar and check whether you have reached an allowance. A temporary rate limit is separate from your monthly allowance.
4. Check whether a workspace setting or organization policy turns off the affected feature. For agent availability, see Agents are not available in chat.
5. Reproduce the issue in a new chat session. When you compare two runs, keep the prompt, model, and available tools consistent.

Check customization diagnostics

Use the chat customization diagnostics view when an instructions file does not appear to apply.

1. Open the Chat view.
2. In the Chat view, open the context menu and select Diagnostics.
3. Review the discovered instruction files and any reported errors.

Use the result to choose the next action:

• The file is missing: Verify its file name, location, and the settings that control instruction discovery.
• The file has an error: Fix the reported frontmatter or content error, and then reopen the diagnostics view.
• The file is loaded but not applied: For a *.instructions.md file, verify that its applyTo pattern matches the file you are working on. Also check the References section of the chat response to see which instructions were used.
• A prompt file is missing from the request: Invoke the prompt file manually in chat. Prompt files do not use applyTo, and agents that run on the Agent Host do not use prompt files.

For file locations, applicability rules, and settings, see Use custom instructions. To verify the exact instructions sent to the model, use the Chat Debug view.

Inspect chat interactions

Use the following tools when the basic checks pass but an agent behaves unexpectedly:

Prepare the session before you reproduce the issue:

• For local chat sessions in the main VS Code window, enable the experimental github.copilot.chat.agentDebugLog.fileLogging.enabled Open in VS Code Open in VS Code Insiders setting and reload the window before you use /troubleshoot.
• For sessions that run on the Agent Host, enable the experimental chat.agentHost.agentDebugLog.enabled Open in VS Code Open in VS Code Insiders setting before the activity that you want to inspect. Capture is not retroactive.
• Reproduce the issue, and then select that session when you open the logs or use /troubleshoot.

To analyze an Agent Host session with chat:

1. Open the Agents window.
2. Enter /troubleshoot #session in the chat input.
3. Select the local or remote session that contains the issue.
4. Add a focused question, such as which tool failed and what error did it return?, and send the request.

For detailed procedures and scenario-based guidance, see Debug chat interactions.

View logs for GitHub Copilot

The GitHub Copilot output channels record connection failures, extension errors, and other runtime information. Increase the log level only while you reproduce an issue because trace logging produces detailed output.

1. Open the Command Palette (⇧⌘P (Windows, Linux Ctrl+Shift+P)).
2. Run Developer: Set Log Level and set the GitHub Copilot and GitHub Copilot Chat extensions to Trace.
3. Reproduce the issue.
4. Run Output: Show Output Channels and select GitHub Copilot or GitHub Copilot Chat.
5. Review errors and warnings around the time of the failed action.
6. Run Developer: Set Log Level again and restore the previous log levels.

To switch output channels, use the dropdown on the right side of the Output panel.

Collect network diagnostics

If Copilot cannot connect, collect diagnostics to identify failed connectivity checks and relevant proxy or network configuration.

1. Open the Command Palette (⇧⌘P (Windows, Linux Ctrl+Shift+P)).
2. Run GitHub Copilot: Chat Diagnostics.
3. Review the diagnostic report that opens in an editor.

If a check fails, compare the result with your firewall, proxy, and VPN configuration. A report without failed checks rules out only the connections covered by the report, so continue with the Copilot output channels if the problem persists.

For required domains and network-specific guidance, see Network and firewall configuration for Copilot.

Troubleshoot MCP servers

If an MCP server or one of its tools is unavailable:

1. Open the Command Palette and run MCP: List Servers.
2. Select the server and review its status.
3. Select Show Output and inspect errors from the failed request.
4. Correct the reported configuration or connection problem.
5. Select Restart Server, and then retry the request.

If the server is running but the model does not invoke its tool, follow An MCP tool is not being invoked.

For setup and server-specific checks, see Configure and debug MCP servers.

Report an unresolved issue

If the issue persists, include enough evidence to reproduce and diagnose it:

• Ghost text suggestions: Hover over a suggestion and select Send Copilot Completion Feedback.
• Next edit suggestions: Select Feedback from the next edit suggestions menu in the editor gutter.
• Chat and agent issues: Open Help > Report Issue, select VS Code Extension, and then select GitHub Copilot Chat.

Include the VS Code version, clear reproduction steps, the expected and actual result, and the relevant diagnostic output.

Related resources

• Debug chat interactions
• GitHub Copilot frequently asked questions
• Troubleshooting GitHub Copilot