AI assistant

The AI assistant is a chat panel inside a terminal session. Ask a question about the host, and the assistant proposes commands, runs the ones you approve on that session, and reads the output to answer or suggest the next step.

​

Open the assistant

In a connected terminal session, click Ask in the bottom-right corner of the terminal. The panel opens over the right side of the focused pane. Drag its left edge to resize it, and click Close assistant to close it. Each session has its own conversation. The assistant is not available in a Files pane.

​

Ask a question

Type a question, or pick a starter such as Why is the disk filling up?, and press Enter. Use Shift+Enter for a new line. The assistant answers and proposes commands as numbered steps. Each step shows what it is for, the command, and a risk dot: low, medium, or high. Nothing runs until you approve it:

• Run runs the command on the session. The output goes back to the assistant, which continues on its own.
• Edit lets you change the command before you run it.
• Reject skips the step.
• Run all appears when two or more low or medium risk steps are waiting.

Safety checks:

• A High risk step needs a second click on Confirm.
• A step that matches a destructive pattern is marked Flagged. Click Run anyway, then Confirm, to run it.

Commands run on the session in the background, with their output captured for the assistant. They are not typed into your terminal. After 5 automatic follow-ups in a row, the assistant pauses until you send a new message. Click Reasoning to see the model’s reasoning, Stop to interrupt a reply, and Clear conversation (trash icon) to start over.

​

Include terminal output

The assistant does not see your terminal unless you attach it. Click Terminal under the message field to send recent terminal output with your next message. If your message mentions something like this error or the output above, the panel offers Attach the terminal output?. Termique scrubs passwords, tokens, and keys from the output before sending it. Your message then shows + N lines of terminal output. Click it to see exactly what was sent.

​

Auto-run

Turn on Auto-run under the message field to let the assistant run steps without asking, for the current session. It is off by default. With Auto-run on:

• Low and medium risk steps run on their own, but only if Termique also classifies the command as read-only. Termique checks the command text on your computer and does not rely on the model’s risk label. Anything it cannot confirm as read-only still waits for you.
• High risk and flagged steps always wait for you.
• Steps run on the hosted provider only. With a custom endpoint, every step waits for you.

Steps that ran this way are labeled auto.

​

Inline suggestion

For a single command, press Cmd+I on macOS or Ctrl+Shift+I on Windows and Linux in a terminal. Type what you want to run under What do you want to run? and press Enter. Termique shows one command with its risk. Click Run to send it to the terminal and run it. High risk and flagged commands need a second click. If the task needs more than one command, click Open AI Chat to continue in the assistant. You can change the shortcut in Settings > Shortcuts.

​

Context

With each message, the assistant receives:

• The host’s label and username.
• The output of commands it ran in this conversation.
• Terminal output, only when you attach it.
• Relevant memories, on the hosted provider.

​

Memory

On Pro, you can save notes the assistant keeps in mind in later conversations, such as how a host is set up or how you like things done.

• Hover an assistant reply and click Remember to save it as a memory.
• Click Memory at the top of the panel to see, add, and delete memories. Under Save to, choose This host or Global.

Each memory can hold up to 4000 characters. With each message, the hosted assistant adds up to 3 memories that match it, preferring ones for the current host. Memories are saved to your account and also appear on the mobile app.

​

Providers

Choose a provider in Settings > AI assistant, then click Save.

​

Hosted

The default. Requests go through the Termique API backend. No API key is needed. Only messages you send count as prompts. The assistant’s automatic follow-ups after a command runs do not. Incident investigation messages count toward the same daily prompts. Limits reset at midnight UTC. The message field shows how many prompts are left, for example 12 left. Hover it to see when the limit resets.

​

Custom endpoint

Use your own OpenAI-compatible API, for example your OpenAI account, Ollama, or another provider.

1. Open Settings > AI assistant and select Custom endpoint.
2. Enter the API base URL, the Model, and the API key.
3. Click Save.

The app sends requests directly to your endpoint, not through the Termique backend, and Termique does not meter them. The API key is encrypted with your DEK and stored on this computer. It is not synced, so enter it on each computer. With a custom endpoint:

• Auto-run does not run anything.
• Memories are not added to requests.
• To attach terminal output, first turn on Send terminal output to this endpoint. Termique scrubs secrets but cannot vet your endpoint.
• Incident investigations still use the hosted provider.