> ## Documentation Index > Fetch the complete documentation index at: https://www.rumus.ai/docs/llms.txt > Use this file to discover all available pages before exploring further. # Smart autocomplete > Inline command suggestions that learn from your shell history and the current context. As you type a command in any terminal tab, Rumus suggests a continuation inline. Hit **→** or **Tab** to accept it; ignore it and just keep typing if it isn't what you wanted. It runs locally, it's on by default, and it doesn't get in your way. ## What it suggests The suggestion engine pulls from a few sources, ranked by recency and relevance: * **Your shell history** — commands you've run before in the current shell, weighted by how recently you used them. * **Filesystem context** — file and directory names in your current working directory. * **Common command structure** — sensible flag completions for tools the suggester recognizes. It's not a model call. The local engine is fast (sub-millisecond for typical inputs) and works offline. ## Accept, partial-accept, or ignore * **Accept the whole suggestion** — press `→` or `Tab`. * **Accept word-by-word** — some platforms support a partial-accept gesture (Alt + →); check your binding in [Settings → Keyboard](/docs/customization/keyboard). * **Ignore it** — keep typing. The suggestion updates as you go; if you type past it, it disappears. ## Turn it off (or tune it) If inline suggestions clash with your muscle memory, turn them off: **Settings → AI → Conversation → Smart Autocomplete** — toggle it. The same section has knobs for fine-tuning suggestion frequency and aggressiveness, when applicable to your shell setup. ## How autocomplete differs from the chat agent These are two distinct features: | Feature | What it does | How it works | | ---------------------- | ------------------------------------------------------- | -------------------------- | | **Smart autocomplete** | Suggests the next characters of your current command | Local, fast, no model call | | **AI agent** | Holds a conversation, runs tools, plans multi-step work | Model call to a provider | Autocomplete is "better tab completion." The agent is "an AI that does things." Use both. ## In remote (SSH) tabs Autocomplete works the same way in remote tabs — it learns from the *remote* shell's history (when [shell integration](/docs/terminal/shell-integration) is enabled, so Rumus can read it back). On a fresh SSH connection with no history yet, suggestions default to local-context fallbacks until you've typed a few commands. ## In broadcast mode The [multi-terminal broadcast](/docs/terminal/canvas#multi-terminal-broadcast) input bar shows suggestions too. Useful when fanning a familiar command across many hosts — type a couple of characters, accept, send. ## Privacy All autocomplete data — your shell history, the directory contents, the suggestions themselves — stays on your machine. Nothing is sent to a server or model. ## Next steps The chat-side AI that runs commands, reads files, and uses tools. Bind the accept-suggestion key to whatever feels natural.