AI Assistant
An AI assistant that reads your setup and proposes changes for you to approve.
- Ask about your hosts, snippets, fleets, alerts, automations and more
- Proposes changes instead of making them, and nothing runs until you approve it
- Can run safe read-only commands like df and uptime if you allow it
- Ask about a host right from its terminal
- Works with Ollama, Anthropic, OpenAI, Google Gemini and any OpenAI compatible endpoint
- Run Pi, OpenCode, Claude Code or Codex as a coding agent on an SSH host, using your own provider
- One AI assistant, floating on every page or full screen in the AI tab, that manages Termix as you through the bundled Termix CLI and opens views for you
- With the ai.ops.use permission, it also runs shell commands across the hosts, folders, tags or fleets you select
- Auto-ops: automations start the assistant to investigate and report, or propose a plan you approve, with ready-made presets
- Asset inventory: a read-only collector maps hosts, Proxmox guests, containers, listening ports and connections into zones, daily and on host changes
- Asset graph: a model annotates each asset's role, importance and business system, with your edits locked, and a graph view shows what fails when an asset does
- The assistant in the panel bar beside any SSH terminal
AI Assistant brings an assistant that knows your Termix setup. Ask it about your hosts, snippets, fleets, alerts and more, or ask it to change them. It works as you, with your permissions, through the Termix CLI that ships inside the plugin. See AI assistant.
It works with a model you bring: Ollama on your own network, Anthropic, OpenAI, Google Gemini, or anything with an OpenAI compatible API. Nothing is sent anywhere until you add a provider and send a message.
Turn it on
The assistant is off for everyone until an admin allows it, and then off for each person until they turn it on.
- An admin opens Settings, AI Assistant and turns on AI assistant.
- Each user turns on Enable the AI assistant on the same page, or just opens the assistant.
- Add a provider under Providers: pick the type, give it a name, and fill in the address, API key and default model.
- Open AI from the sidebar, or press Ctrl+J on any page, and ask something.
Self-hosted models
Termix blocks requests to private addresses by default. To use Ollama or another model on your own network, an admin lists its host under Allowed private AI hosts, one per line. localhost, 127.0.0.1, ::1 and host.docker.internal are allowed out of the box.
The assistant
One assistant, in two views that share their conversations: a floating panel on every page, and the full-screen AI tab. It changes things without asking first, and with the ai.ops.use permission it can also run shell commands across hosts you select. See AI assistant.
Auto-ops
With the Automations plugin, automations can start the assistant on their own: on a schedule, when a disk fills up, when a host goes down. It investigates read-only and writes a report, or proposes a plan that runs only after someone approves it. Presets for a daily patrol, disk cleanup and host down triage get you started. See Auto-ops.
Asset inventory
Assets, in the AI tab, collects every host you can connect to with one read-only script: listening ports, connections, services, containers, Proxmox guests and cloud metadata. It sorts assets into zones by built-in and your own CIDR rules, marks hosts that sit in more than one zone, and links guests and containers to where they run. See Asset inventory.
In the terminal
Turn on Enable AI Assistant in a host's settings to get an assistant panel next to its terminal. It knows which host you are on and reads your setup through a fixed set of tools.
Open it from the terminal toolbar button, which also hands it what is on screen, from the tab menu, or from the edge of the terminal. On Termix 26.10.3 and later it sits in the panel bar with the other panels, so you can resize it, move it to either side or the bottom, and turn it off for a host with Panel beside the terminal.
| It can read | |
|---|---|
| Hosts | Names, addresses and settings. Never passwords or keys. |
| Snippets, fleets, automations | If those plugins are on. |
| Workspaces, homepage items, the network map | If those plugins are on. |
| Notification channels | Names and types only. |
| Your command history | From the SSH terminal. |
To change something, it proposes it. A card shows exactly what would happen, and nothing runs until you press Approve; commands it suggests can run in your open session. Approving needs the Apply proposals permission.
Turn on Allow read-only diagnostic commands and it can run safe commands without asking each time, to answer things like "how full is the disk". Only a fixed list runs this way: df, du, free, uptime, uname, ps, top, ip, ss, lsblk, cat (only system files under /proc and /sys), ls, systemctl status, journalctl, docker ps and docker logs, and a few more. Anything else is proposed first.
Coding agents
The plugin can also start a coding agent, like Claude Code, Codex, OpenCode or Pi, on an SSH host and drive it from Termix. See coding agents.
Permissions
| Permission | What it allows |
|---|---|
ai.use | Use the assistant, in the AI tab, the floating panel and the terminal. |
ai.manage_providers | Add, edit and remove providers and their keys. |
ai.apply_proposals | Approve the changes the terminal assistant proposes. |
ai.agents | Start coding agents on hosts. Admins only by default. |
ai.ops.use | Let the assistant run shell commands on hosts, and use auto-ops. Admins only by default. |
ai.usage.view_all | See every user's token usage. Admins only by default. |
ai.assets.view | See the asset inventory. Admins only by default. |
ai.assets.manage | Collect assets, edit zone rules and the schedule. Admins only by default. |