How to use AgentVera: first project, agents and grid

How to use AgentVera step by step: add a project folder, start Claude Code or Codex agents, arrange the grid, follow their status and sign in only for Pro.

Solviera Teknoloji 8 min read

To use AgentVera, install the coding agent CLIs you want, add a project folder, start one or more agents from New agent, and work with them side by side in the project’s grid. Each agent is the real CLI, such as Claude Code or Codex, running in its own terminal pane with its own login. AgentVera adds the parts around it: a per-project grid, a message box with a queue, status at a glance, per-turn checkpoints, code review and account handling.

This guide takes you from an empty window to a few working agents. It covers what to install first, how the New agent dialog works, how to arrange the grid, how sign-in works for the CLIs and for AgentVera itself, and the habits that keep parallel agents out of each other’s way.

What you need before you start

AgentVera is a desktop app for macOS, Windows and Linux. You can get it from the download page. It does not ship its own model. It runs the agent CLIs already on your computer.

You needWhy
At least one agent CLI, such as Claude Code or CodexAgentVera starts the real CLI for every agent
A login for that CLIAgents use the CLI’s own sign-in and plan
A project folder, ideally a Git repositoryCheckpoints, worktrees and code review build on Git
An AgentVera account (optional)Only for Pro and Team features

If you have never used Claude Code, How to use Claude Code covers installing it and logging in. AgentVera supports 18 agent CLIs, including Claude Code, Codex, Gemini CLI and GitHub Copilot CLI, so you can start with whichever you already pay for.

Step 1: add your first project

Open AgentVera and click Add project folder in the top bar. Pick the folder of a repository you work on. The project appears on the Projects home screen and in the top bar, and you can jump between projects from there or with the command palette (⌘K, Ctrl K on Windows and Linux).

Each project has two views:

  • Workspace: the grid of agent panes, where you spend most of your time.
  • Details: information about the project around the agents.

If you just want to ask a quick question before setting up a project, the home screen has a quick chat. Those chats run in the app’s own folder and don’t touch your projects.

Step 2: start your first agent

Click New agent in the sidebar, or press ⌘N. The dialog says what it wants: pick a tool, then give the agent a name and, if needed, a role.

  1. Tool: choose the CLI, for example Claude Code or Codex. A tool AgentVera can’t find on this computer is marked Not installed.
  2. Agent name and Color: a short name such as “Backend” or “Reviewer” makes the grid easier to scan.
  3. Runs on: leave it on This computer for now. You can also run an agent on an SSH server.
  4. Account: the login the agent uses. The default is the one already on this computer. The field shows Signed in or Signed out.
  5. Workspace: the main folder or one of the project’s Git worktrees.
  6. Role and instructions: optional. For Claude it is added to the system prompt, and for Codex it is added as developer instructions.
  7. Click Start agent.

The dialog has more options, including context profile, browser, project map, MCP servers, Docker container, token budget and owned paths. You can leave all of them at their defaults for a first run and come back later. Token savings explains context profiles, and Docker explains running an agent in a container.

The agent opens in a pane. That pane is a real terminal running the CLI, so everything the CLI does in a normal terminal works here too, including its slash commands. Under it is a message box with prompt history, drafts and a queue (⌘Enter queues a message while the agent is still busy).

Step 3: sign in the agents

Agents don’t log in through AgentVera. They use the CLIs’ own logins, so if claude or codex already works in your terminal, the default account is ready. AgentVera tags that default account this computer.

If an account is signed out, the New agent dialog warns you: “This account is signed out. Sign in from Settings → Accounts, or the agent will ask you to sign in on first launch.” You can follow either path.

To use more than one Claude or Codex account, open Settings → Accounts, click Add Claude account or Add Codex account, give it a name and choose Add and sign in. Each extra account gets its own config folder, and you pick which account an agent runs with when you create it. How to switch Claude Code accounts walks through this in detail.

Step 4: add more agents and arrange the grid

One agent is useful. Two or three in parallel are where AgentVera earns its place. Create a second agent the same way, and give each one a clear, separate job.

The grid lives in Screen splitter (⌘⌥L). Pick the number of columns and rows, from 1×1 up to 5×4, or a ready-made layout such as “Top 2 · Bottom 1”. Drag the dividers to resize panes, and use Equalize pane sizes to reset them. Each project remembers its own layout.

ActionHow
Swap two panesDrag a pane onto another
Maximize a paneDouble-click its title, or ⌘⇧Enter
Move focus between agents⌘⌥ + arrow keys
Jump to a pane⌘1 to ⌘9
Open the command palette⌘K
Open the screen splitter⌘⌥L

On Windows and Linux, use Ctrl in place of ⌘ and Alt in place of ⌥. You can change every shortcut in Settings. A pane can also open in a modal or in a separate window if you want one agent on a second screen.

When several agents share one folder, they can overwrite each other’s files. Pointing each agent at its own Git worktree in the Workspace field avoids that. Git worktrees for parallel agents explains the setup, and How to run multiple Claude Code agents covers splitting the work.

Step 5: follow what each agent is doing

Every agent shows a status: Working, Waiting for you, Needs approval, Idle, Asleep, Stopped or Error. The status bar counts them, for example “2 working” or “3 agents need approval”, so you can see what needs you without opening each pane. Press ⇧⌘A to go to the next agent waiting for an approval.

When you step away, desktop notifications tell you when an agent finishes, fails or waits for approval. You can also have them sent to your phone. See notifications and approvals.

Step 6: check the work and undo a bad turn

Two Free features help you trust what parallel agents produce.

  • Code review: when you turn it on in the project settings, a Claude model you choose reviews each agent’s code when its turn ends. Findings come with a severity and a file:line link, and can optionally go back to the agent. See code review.
  • Checkpoints: at the start of each turn, AgentVera snapshots the agent’s working folder. The Turns button in the pane header lists the turns and the files each one changed, and Roll back this turn takes the folder back. How to undo Claude Code changes covers this step by step.

Step 7: sign in to AgentVera only when you need it

Most of what this guide covers is on the Free plan, with no account. Pro and Team add features such as flows, night shift, auto approval and automatic account switching. To use them, choose Sign in to AgentVera in the app. You can sign in with the browser (Google sign-in and two-factor authentication are handled there) or with email and password. The pricing page lists what each plan includes.

A first session that works

If you’re not sure what to try first, this is a simple setup that shows the value quickly:

  1. Add a repository you know well.
  2. Start a Claude Code agent in a new worktree and ask it for a small, well-scoped change.
  3. Start a second agent (Codex, or another Claude Code) in its own worktree on an unrelated task.
  4. Set the grid to two columns and keep both visible.
  5. While they work, read the diffs, answer approvals and queue your next message.
  6. When a turn goes wrong, roll it back from Turns instead of arguing with the agent.

Tips for the first week

  • One job per agent. Name agents after their job. A clear role in Role and instructions saves you from repeating yourself.
  • One worktree per agent. It is the easiest way to stop agents from overwriting each other.
  • Keep the grid small at first. Two or three agents you actually read beat eight you only skim.
  • Use the palette. ⌘K finds almost every action, with its shortcut next to it.
  • Watch the limits. The status bar shows each account’s 5-hour and weekly usage. Claude Code usage limits has tips for making them last.
  • Switch the language if you like. The interface is in English and Turkish, with light and dark themes, under Settings → Appearance.

Wrapping up

Getting started with AgentVera takes a few minutes. You add a project, start an agent with the CLI you already use, and add a second one when you have a second task. Everything else, such as worktrees, checkpoints, review, extra accounts and phone notifications, builds on that same grid. You can add each one when you need it.

Questions

Do I need an AgentVera account to start?

No. Projects, agents, the grid, Git, MCP, Docker, SSH, code review and token savings are on the Free plan and need no account. You sign in to your AgentVera account only to unlock Pro or Team features.

Does AgentVera replace Claude Code or Codex?

No. AgentVera runs the real CLIs. Each agent is a Claude Code, Codex, Gemini CLI, Copilot CLI or other CLI process in its own terminal, and it uses that CLI’s own login and settings.

What if the tool I pick shows Not installed?

AgentVera could not find that CLI on this computer. Install it the way its maker documents, make sure the command runs in a normal terminal, then open the New agent dialog again.

How many agents can I run in one project?

The Free plan has no agent limit. The grid shows up to 5 columns by 4 rows of panes per project, and each project remembers its own layout.

Which systems does AgentVera run on?

macOS, Windows and Linux. It updates itself from agentvera.dev, and agents return to their conversation the next time you open the app.

Bring your agents to one desk.

Download AgentVera for free; your installed CLIs are ready to go.

More posts