AI Not Responding

No AI provider is connected yet. Open AI Keys from the left sidebar, under TOOLS, then pick one of three ways to connect:

• Sign in with ChatGPT: on the first Quick connect card, if you already have a ChatGPT account. No API key.
• Connect with OpenRouter: on the second card. A free OpenRouter account is enough - CtrlOps picks a free model for you.
• Add credential: in the API credentials section, to use your own key from OpenAI, Anthropic, Google, or a custom OpenAI-compatible endpoint.

See the AI Terminal page for the full walkthrough.

The API key isn't being accepted by the provider. Causes, in order of likelihood:

• The key was copied with extra whitespace. Re-copy from your provider's dashboard and re-paste.
• The key was revoked or rotated. Generate a new one in the provider's console.
• The key has the wrong permissions / scope. OpenAI keys need access to chat completions. Anthropic keys need messages access. Most personal keys have everything by default.
• You're out of credits. Many providers freeze keys when the prepaid balance hits zero.

Fix: Left sidebar → TOOLS → AI Keys → click Edit on the credential → paste the new key → click Test connection before saving.

The provider's API isn't returning a list of models. Three things to check:

• API key first. A wrong key shows the same symptom. Click Test connection, if it says Unauthorized, fix the key.
• For OpenAI Compatible providers: the Base URL is required and must include the version path, for example https://api.your-provider.com/v1, not just https://api.your-provider.com.
• Network issues between you and the provider. Check by visiting the provider's dashboard in a browser, if that loads, the network is fine.

You've sent more requests in a short window than the provider allows. The fix depends on your provider:

Switching to a cheaper model from the dropdown also helps, because cheaper models often have higher rate limits.

A few things to try, in order:

• Switch to a faster model from the dropdown at the bottom-left of the AI panel. Try claude-haiku, gpt-4o-mini, or gemini-flash.
• Reduce the conversation length. Click New in the AI panel header to start a fresh chat. Long histories make every response slower.
• Lower the Max AI reasoning steps in the panel footer. The AI may be running many internal steps before answering. Try 20 to 50.
• Check the provider's status page if responses are abnormally slow across many sessions.

The AI hit the Max AI reasoning steps cap mid-task. Two fixes:

• Raise the cap. The slider in the AI panel footer goes up to 200. Bump it for complex tasks.
• Break the task in half. "Set up nginx + issue an SSL cert + start the app" is three tasks. Ask each one separately, then chain them.

A few things sharpen the responses:

• Be specific. "Apache shows 'Address already in use' on port 80" gets a better answer than "my server is broken".
• Paste the exact error. The AI is excellent at decoding cryptic error text but can only do that if it sees the text.
• Mention the OS. Ubuntu, CentOS, Amazon Linux, all have slightly different commands. Saying so up front avoids generic guesses.
• Start a new conversation when you switch topics. Long conversations drift.
• Try a stronger model (claude-opus, gpt-4o, gemini-pro) for hard or unusual problems.

The terminal session may have disconnected. Look at the bottom-left for a "Disconnected" alert, click Reconnect, then ask the AI to retry.

If the terminal is connected but Run still does nothing, switch to another tab and back to refresh the panel. As a last resort, right-click the terminal and choose Reset Session.

Auto-Run only auto-approves commands the AI proposed itself in the same conversation. Some safety-critical commands, particularly those that delete files or stop services, may still surface a confirmation. That's intentional.

If you want zero prompts at all, the AI panel's Auto-Run toggle is the only switch, there is no separate "safety level" setting in the product today.

Reset by switching tabs:

• Click the File Manager tab in the sidebar, then back to AI Terminal.
• Click Ask AI in the header.

If it's still stuck, restart the app. The chat history is preserved as long as you don't click New.

A few common cost traps:

• Heavy models on every question. GPT-4 Opus / Claude Opus / Gemini Pro are 5 to 20 times more expensive per token than their smaller siblings. Switch to a small model for routine questions and reserve the big ones for hard problems.
• Auto-Run on with chains of commands. Each command's output gets fed back to the AI, which decides on the next command. Twenty hops add up.
• Web Search on by default. Web search adds tokens. Toggle it off when you don't need it.
• Long-running conversations. Old messages in the same conversation are re-sent on every new question. Click New when starting a different topic.

You can see exact costs in your provider's dashboard. Set a monthly budget cap there if you want a hard limit.

The AI sometimes returns plain code without specifying a language. To force highlighting, ask it to "format the answer as bash" (or python, sql, nginx, etc.). The next response will use that language hint.

Highlighting is supported for: bash, javascript, typescript, python, php, sql, yaml, json, nginx, and most other common languages.