Actovia

Built-in tools

What the model can do on its own, and how to switch each tool off.

Actovia's model does not just talk: it can look things up, work with files, run commands and draw things, using the tools below. It uses them without asking first and chains them until the task is done — searching, then reading a page, then writing a file. It stops to ask only when something is genuinely unclear, or before anything on its short list of hard stops: deleting or overwriting outside the chat's folder, sudo, installing system-wide, git push, or anything sent or paid for on your behalf.

Turning tools on and off

Settings → Built-in Tools lists every tool with its switch, grouped under Web, File System, AI & Data and Maps. A tool that is off is never offered to the model. A tool that needs a key or has options carries them under its row (Configure folds them away); while it is on and its key is missing, a red hint says so. Getting API keys says where to get each one.

Settings → Built-in Tools: rows with switches, Web Search’s panel open with the saved Brave key, Country, Languages and Max results.
Settings → Built-in Tools: rows with switches, Web Search’s panel open with the saved Brave key, Country, Languages and Max results.
ToolWhat it doesNeeds
Web SearchSearches the web, with citations (details)Brave Search API key
Web FetchReads a page, a doc, an API or a GitHub file—
WeatherCurrent weather and a seven-day forecast (details)—
Map ItineraryPlaces, routes and multi-day trips on one mapGoogle API key
TerminalRuns shell commands (details)—
Read / Write / Edit File, List Directory, Find Files, GrepWork with files—
Image GenerationMakes pictures (details)OpenAI API key
Knowledge BaseLooks things up in your knowledge baseA LightRAG server
Update MemoryCorrects, adds or removes memories—
RecallReads back an earlier search result or tool output in full—

A few more abilities are always there or are switched on elsewhere: artifacts, Computer Use (experimental, off by default), Deep Research, and the tools of your MCP servers and skills.

Recall

A tool's full output matters for the step that called it. In later turns the model is sent a short digest instead — a line per search result, the last lines of a command — which keeps long chats fast and cheap. When you come back to it ("what exactly did the third source say?", "show me that command's whole output"), the model uses Recall to read the stored original back, word for word, without searching again. Weather is the exception: an old forecast is stale, so the model checks again.

Asking before secrets

Some files and commands touch credentials — SSH keys, cloud logins, browser profiles, .env files outside the chat's folder. The model can still reach them, but only after you say yes. See Approvals.