Docs
Using Agent V
How each part of Agent V works, from the first launch to a merged pull request. New to it? Start with Getting started, then follow the use cases.
Start
- Getting startedInstall Agent V, work through the three setup steps, and hand the agent its first task.
- Models and keysThe model providers Agent V supports, how to add an API key, where keys are kept, and how to use local or custom endpoints.
- WorkspacesHow Agent V uses a folder as a workspace, the home workspace, recent folders, and running a task in its own git worktree.
Tasks
- Writing a taskThe New task page, writing a brief, Done when checks, Agent and Ask modes, model and effort, drafts, and dictation.
- The task recordHow to read a task's record, its plan and steps, the receipt line, and how to add instructions to or stop a live run.
- InstancesHow a task splits work across sub-agents called instances, what each one is told, how they are isolated, and what the parent gets back.
- Agent typesDefine a helper once, as a markdown file with a name, a one-line purpose, the tools it may use and its model, and the agent can start instances of it.
- ApprovalsWhat asks for your OK before it runs, the three approval modes, the approval card, Always allow rules, and runs nobody is watching.
- Goals and loopsKeep a task working toward an objective across turns with /goal, and repeat an instruction on a timer with /loop.
- Scheduled tasksRun a brief or an earlier task again on a schedule, as a new task each time, and how that differs from /loop.
- Voice modeTalk with the agent on a task, choose who listens and who speaks, add your own keys, interrupt it, and approve by voice.
Review and ship
- Review and rewindReview what a task changed in the inspector, keep or undo each file, rewind a task to an earlier instruction, and redo or rerun it.
- Git and GitHubWhat needs git, committing from the Changes tab, connecting GitHub, and working with pull requests and failing checks in the PR tab.
- Usage and costWhat the Usage page shows, how Agent V works out the cost of a run, and what "est." and "~" mean.
Extend
- ExtensionsAdd MCP servers, skills, rules and packages from the Extensions page, choose where each one runs, and how MCP tools load.
- Rules and skillsWhich rule files Agent V reads and in what order, rule frontmatter, your own global rules, and how skills load when a task needs them.
- Agent-built toolsHow the agent writes its own tools with build_tool, where they live, why they are not sandboxed, and how their approvals work.
- HooksRun your own commands when a task starts, when you send an instruction, before and after the agent's tool calls, before older work is summarized, when it is about to finish, when a task needs you, and when a turn ends.
- Built-in toolsMoving and renaming files, web reading and search without the browser, reading the browser's console, the workspace's memory, and tools that load only when a task needs them.
Reference
- Keyboard shortcutsEvery keyboard shortcut in Agent V, grouped by what it does.
- AppearanceChange Agent V's skin, colour mode and text size, and load your own CSS on top.
- Data and storageWhere Agent V keeps settings, keys, task records, undo points, logs, worktrees and memory, and what leaves your computer.
- TroubleshootingWhat the app's error messages mean and what to do about each one: keys, providers, MCP servers, the proxy, settings, updates, worktrees and start-up.
- Headless runsRun one task with no window, from a script or CI, with Vyotiq --headless — options, approvals, output and exit codes.
- Notifications and closingWhere Agent V tells you things, notifications on your phone, sound and quiet hours, budget alerts, and what closing the window or quitting does to running tasks.
- TerminalThe Terminal tab — your own shells beside the agent's commands, processes the agent left running, find, copy, clear, rename, other shells, and handing output to the agent.