Getting started

Agent V is a desktop app. You give it a task in a folder on your computer, it plans the work, edits files and runs commands there, and shows you what it did.

Install

Builds are published on the releases page. The download page links to the latest one. The files are named Vyotiq.

System	What to get
Windows	The Vyotiq-<version>-x64-setup.exe installer, or -arm64-setup.exe on a Windows on Arm PC. It asks whether to install for you only (the default) or for everyone on the computer, and lets you pick the folder.
macOS	The .dmg for Apple silicon (arm64) or Intel (x64).
Linux (x64 or arm64)	An AppImage, a .deb or an .rpm, each in an x64 and an arm64 build.

Updates

Agent V checks GitHub for a new version when it starts and every 6 hours after that. Nothing downloads or installs on its own. You can turn the check off in Settings, “About”, under “Updates”: “Check automatically”. “Channel” there picks “Stable” (the default) or “Beta”, which also offers pre-releases: newer features, rougher edges. Going back to Stable never downgrades; the next stable release newer than yours is offered.

First launch: “Set up Agent V”

The first screen is “Set up Agent V”: “Three things, then hand it its first task. Everything here can change later in Settings.”

1. “Connect a model provider”

The step checks the provider that is set, Ollama by default. When it has nothing to connect with, the step offers what it can right there:

• “Found on this computer”: an API key already set in your environment (ANTHROPIC_API_KEY, OPENAI_API_KEY, OPENROUTER_API_KEY, GEMINI_API_KEY or GOOGLE_API_KEY, DEEPSEEK_API_KEY, GROQ_API_KEY, XAI_API_KEY, MISTRAL_API_KEY) with “Use this key”, which copies it into the key vault; and Ollama or LM Studio running on this machine, with how many models it has and “Use it”.
• Paste a key: pick the provider, paste its key and click “Save”.

“More providers in Settings” (or “Change”) opens Settings, “Providers”, for every other provider and option.

The step only counts as done after Agent V reads the provider’s model list successfully, so a wrong key or an unreachable server shows up here, not in your first task. If the check fails you get the reason, with “Check again” and “Choose a provider”.

Ollama needs no key when it runs on your own machine. See Models and keys for every provider.

2. “Open a workspace”

“The folder the agent works in. Its file tools stay inside it.”

Click “Choose a folder…” or drop a folder onto the step. Folders you opened before appear under “Recent” (up to five). See Workspaces.

3. “Decide what needs your OK”

“You can change this per workspace.” Pick one:

Choice	What it means
“Edits and commands”	“Recommended. Reading is free; changing things asks first.” This is selected when setup opens, unless you already picked another mode in Settings.
“Every tool”	“Even reads and searches ask.”
“Nothing”	“For runs nobody is watching. MCP tools, tools the agent writes and risky commands still ask.”

With MCP protection off in Settings, “Agent”, MCP tools no longer ask. Approvals explains each mode in detail.

When the first two steps show done and you have picked a choice in step 3, click “Start your first task”.

Your first task

The button opens the “New task” page for the workspace you chose. Later, “New task” at the top of the navigator opens it too (Ctrl+N, ⌘N on macOS).

1. Describe what you want in the brief. Write the result you expect, not just the topic.
2. Optionally add checks in the brief’s “Done when… a command passes, a file exists” line: type one and press Enter. The agent marks each check met or not met, with its evidence. A check it leaves unmarked shows “Not checked yet”.
3. Click “Start task” or press Ctrl+Enter (⌘↵ on macOS).

The task opens as a record. You watch each step as it happens: files read, edits made, commands run. When the agent needs your approval, a card appears in the record with the choices. A receipt line shows the time taken, how many checks were met, the tokens used and, when it is known, the cost (“~” means estimated).

Next, follow the use cases in order. The first, Understand a codebase you don’t know, only reads your code, so it is a safe place to start. For the details of each part, read Writing a task and The task record.