AgentVera is a desktop app that runs Claude Code, Codex, Gemini, Copilot, Cursor and 13 more coding agents side by side, per project. Each agent runs in a real terminal with your own install and your own account.
Start on the free plan. For macOS, Windows and Linux.
Drag an agent here from the left
~/code/acme-web
Add agents and logic nodes from above, then connect them.
No steps yet. They show up when an agent finishes a turn or a Start node runs.
Pick a ready-made server, or enter your own command or address.
Browser
Docs
Development
Mobile
Data and memory
Product and team
Design
This is a demo: the real interface with pretend agents. Click around, build a flow, add an MCP server, start a new agent. The demo is easier on a desktop screen, but you can tap around here too.
AgentVera doesn’t bring a new model or a new account. It opens the CLIs on your machine and uses their logins as they are. Your subscription and API keys stay where they are.
18 coding agents and a plain terminal
Marked tools pick up AgentVera’s MCP servers and browser on every launch. Account chains and context profiles work with Claude Code and Codex.
Try adding an agent in the windowNo API key needed to start. AgentVera finds the CLIs on your machine and uses them as they are; you sign in with an AgentVera account for Pro features.
Pick a folder from the projects screen; a Git repository is recommended. AgentVera also finds CLIs installed through Homebrew, NVM, Volta and asdf.
Choose a tool under New agent and give it a name, color, model, effort and role. Lay out anything from 1×1 to 5×4 and drag agents into panes.
Wire agents together with flows, turn on automatic review at the end of each turn, and approve waiting steps from your desk or from Telegram.
Click a project and its agents open in the layout you left. Switch to another project and they keep working in the background.
Quit and reopen the app and every agent except plain terminals returns to the same conversation from its CLI’s history.
Move between agents with ⌘⌥ and the arrow keys, jump to a pane with ⌘1–9. Every shortcut can be remapped.
Queue a message while an agent is busy, or send one message to the agents you pick.
Run time, agent hours, tokens and model mix come from your local conversation logs.
38 hagent hours this week
2.4 Mtokens
126prompts
Wire agents together on a canvas. When the source agent finishes its turn, its last reply is dropped into a template and sent to the target. If the target is busy, the flow waits for its turn to end, and turn limits stop loops.
Nodes: start, condition, wait, transform, notify and human approval.
Build one in the window0 9 * * 1-5 Send an agent a message on a cron schedule, like updating dependencies every weekday at 09:00. If the agent is busy, wait, send anyway or skip.
Your databases, team boards and account now live in the same app. Agents work overnight, pass safe requests without asking and reach servers under your rules.
Connect to PostgreSQL, MySQL/MariaDB and SQLite; edit tables, write SQL, see the ER diagram. If you like, agents reach the database at the level you choose and never see the password.
Navbar → Database
Run the project with your team on a kanban board. Hit “Assign to Claude” on a card; the agent takes the task, posts its reply to the card when its turn ends and moves the card to review.
Navbar → Kanban
Queue tasks in the evening; overnight each agent works in its own worktree, runs the tests and leaves its work as a draft pull request. A report arrives in the morning.
Home → Night shift
Safe requests in Claude Code, such as reading tools, read-only commands, tests, lint and builds, are approved without asking; push, delete and irreversible commands always come to you.
Settings → Auto approval
Open a server to agents and choose the level: read-only, ask on change or free. Allow and deny lists apply to every command, every command is logged, and the agent never sees the password or key.
SSH → server → Agent access
Write the goal and pick a tool for each stage; three agents work in turn. A stage only moves on once the agent hands its work over, and the plan waits for your approval before the builder.
⌘K or Details → Flows
Send a selection from the terminal, editor, diff view, browser console, Docker logs or an API response to an agent with one shortcut (⌘⇧L). Where an error is caught, a “Fix” badge appears.
⌘⇧L or right-click
Your project’s endpoints are found from OpenAPI files and from routes in frameworks such as Express, Next.js, FastAPI, Django, Spring, NestJS, Go and Laravel. Send a request, see the response, have an agent fix it if needed.
Navbar → API
Focus a group and see only its agents. In overview mode each agent’s status, last reply and pending approval sit on one card; approve, reject or send a quick message.
Sidebar → View
Line up as many Claude and Codex accounts as you like. When the remaining limit drops below your threshold, or the CLI reports a limit, the agent moves to the next account. The conversation is copied over and the agent carries on.
personal@ Switched work@ Active spare@ Next Pick a context profile per agent. That baseline is re-read on every turn, so the savings compound over long conversations. When context grows, AgentVera runs /compact at your threshold.
Context sent to the model on the first request of a new session. Measured with Claude Code 2.1 and Codex 0.155.
When a turn ends, the code it wrote, either the commits from that turn or all changes, is reviewed by the Claude model you choose. Findings come with a severity and a file:line link, and you can send them back to the agent in one click.
Up to two rounds in a row. You can also review the working tree or a single commit by hand.
Refresh token reuse isn’t detected; a new token is issued before the old one is revoked.
src/auth/session.ts:88 The database transaction isn’t rolled back on the error path.
src/auth/store.ts:41 Unused import.
src/auth/index.ts:3 AgentVera’s MCP manager keeps servers in one list and attaches them when an agent starts. Besides Claude Code and Codex, the other tools that take MCP (opencode, Copilot, Qwen Code, Goose, Amp) get the same servers.
Explore the MCP managerSome of the catalog
20 servers grouped by category: browser, docs, development, mobile, data, team tools and design. Add one in a click, or enter your own command or address.
Give a server to every agent or to the ones you pick, and limit it to one project if you like. The change applies when the agent restarts.
A test lists the server’s tools, prompts and resources, with the approximate tokens each tool adds to the agent’s context. You spot a heavy server before it bloats every turn.
Tokens in environment variables and headers live in your operating system’s keychain. Servers that need OAuth are authorized on first use.
Servers from Claude Code’s and Codex’s own settings are listed with their health; import them into AgentVera, or write AgentVera’s servers back to the CLI configs.
The agentvera-board server is added to every project automatically, so agents can share notes and decisions on one board.
The tools your agents need live in the same app, each with its own shortcut.
Playwright, GitHub, Sentry, Linear, Figma and more. Add from the catalog, test the connection, see the token cost of each tool. Secrets live in the Keychain.
A separate Chrome profile. The agent browses with Playwright while you watch the tab live and step in when you want.
Changes, graph history and file diffs. Claude Haiku writes commit messages in your repo’s style.
Containers, live logs, exec, images and compose, without opening Docker Desktop.
Saved servers, ProxyJump, port forwarding and two-pane file transfer. Run agents directly on the server.
Full-text search across every Claude and Codex conversation, including ones started outside AgentVera. Resume any of them in a new agent.
Turn finished, errors, flows waiting for approval. Approve from your phone or reply to message the agent.
Every action, agent, project and server in one search box.
All of these are things the app does today.
Open a branch per feature with the worktree manager and run an agent in each. Review the diffs GitHub-style in the Changes panel.
Claude writes the plan and the code, the flow carries its reply to Codex, Codex writes the tests. At the end, the flow waits for your approval.
Pick a saved SSH server and folder; Claude Code or Codex runs on the server with the server’s own login.
Schedule a message to an agent with a cron expression, like a dependency update. If the task fails, Telegram lets you know. Runs while the app is open.
Broadcast the same prompt to agents running different tools and read the replies side by side.
You can already run coding agents in terminal tabs. What AgentVera adds is managing the work between them.
| Separate terminal tabs | AgentVera | |
|---|---|---|
| Keeping an eye on agents | Switching tabs one by one | One grid, side by side, with working, waiting and error states |
| Passing one agent’s output to another | Copy and paste | Flows with templates, conditions, approval and turn limits |
| When a limit runs out | Sign out and sign in with another account | Account chains for Claude and Codex, conversation included |
| When you quit | You keep track of session ids yourself | Agents return to their conversations |
| Code review | On request | Automatic at the end of a turn, findings at file:line |
| MCP servers | Each CLI’s own config file | One list, per agent |
| Sign-in and subscriptions | Each CLI uses its own login | Same: each CLI uses its own login |
Sign in with your account, set up your organization, invite your team and run the work on a kanban board together with your agents.
Agents run on your computer with your own CLI sessions. Your code and agent conversations are never sent to AgentVera servers; with team kanban, only the cards, comments and attachments you put on a board are shared.
Every session is a real pseudo-terminal opened with node-pty.
Commands start as an executable plus an argument array.
Passwords, tokens and MCP secrets live in your operating system’s keychain.
The UI has no Node or file system access; everything goes through typed IPC.
It never leaves the project root and doesn’t follow symlinks.
Records are written atomically to a single file with 0600 permissions.
Every core feature is on the free plan. Automation and team features such as flows, night shift, auto approval, the database manager and team kanban come with Pro and Team.
Free
$9.99/ month
$6.99/ month / seat
At least 3 seats
Prices are in USD and include VAT. Customers billed in Türkiye pay the TRY equivalent at the central bank (TCMB) selling rate on each charge day.
No. AgentVera is a desktop app that manages the coding agents installed on your computer (Claude Code, Codex, Gemini CLI, Copilot, Cursor, Antigravity, opencode, Hermes, Qwen Code, Kimi, Kiro, Amp, Droid, Goose, Vibe, Crush, Aider and Cline) in one interface. Models and billing come from each agent’s own account.
Any CLI runs in a Terminal agent. The listed tools also take settings like model, effort and role from AgentVera.
Yes. Each agent uses its CLI’s own login. If you signed in with a Claude or ChatGPT plan, AgentVera uses that session and doesn’t ask for an API key.
Core features are free and agents are unlimited. Flows, night shift, auto approval, automatic account switching, the database manager and team kanban come with Pro and Team; current prices are on the Pricing page.
AgentVera runs on macOS, Windows and Linux. Setup is the same everywhere: download, pick a project folder, open your agents.
Yes. When you create an agent, pick a saved SSH server and folder. Claude Code or Codex runs on the server with the server’s own CLI login and shows up in AgentVera like a local agent.
Terminals are shut down cleanly. Next time you open the app, every agent except plain terminals returns to its conversation from its CLI’s local history.
Project and agent records, your code and conversations stay on your computer (~/Library/Application Support/AgentVera on macOS); passwords live in your operating system’s keychain. When you use accounts, organizations and team kanban, your account details, device record and the cards, comments and attachments on boards are stored on the AgentVera server.
On Pro and Team plans, for Claude Code agents: reading tools, read-only commands, test, lint, typecheck and build commands, and read-only git commands. Push, delete, irreversible git commands, output redirection and anything not on the lists come to you. Turn it off or change the lists in Settings → Auto approval.
Yes. Tasks run while the app is open and the computer is awake. Each task runs in its own worktree and branch, the agent doesn’t push, finished work is opened as a draft pull request and a report arrives in the morning.
Yes. Set up an organization with your AgentVera account and invite your team; on Pro and Team plans you get the team kanban. Agents still run on each developer’s own computer; the board shares cards, comments and assignments.
Not for the core features: download, pick a project folder, open your agents. Pro and Team features such as flows, night shift, auto approval, the database manager and team kanban need you to sign in to the app with your AgentVera account.
No. Your voice is transcribed on your computer with the whisper.cpp build and model that ship with the app; nothing is sent anywhere.
The interface is in Turkish. There is an English option, but it’s partial for now; untranslated text stays in Turkish.
Yes. AgentVera snapshots the working folder to a hidden git ref at the start of every turn. From the Turns list you see the files a turn changed and undo it, without touching your branch or stash.
Add any of the 20 ready-made servers in the catalog (Playwright, GitHub, Sentry, Linear, Notion, Supabase, Figma and more), or define your own over stdio, HTTP or SSE.
Download, pick a project, open your first agent. Your installed CLIs are ready to go.