Desktop cockpit for AI coding agents

Run many agents at once. Keep the decisions.

ove.run runs eight agent CLIs side by side, gives them shared project knowledge, and checks their work against quality gates before you accept it.

Download
ove.run session grid with several AI coding agents running side by side, each showing its live status.
The problem

One agent is a tool. Five is a management job.

Each agent runs in its own terminal. It asks questions you miss, starts without what the last one learned, and reports done before the tests pass.

ove.run puts every agent, every question and every check in one window.

Agents

Eight agent CLIs. One grid.

Claude Code, Codex, Gemini, Copilot, Cursor, OpenCode, Amp and Antigravity, plus a plain terminal. Run them side by side, in any mix.

  • Multi-pane grid of terminals and chats, with live status per session
  • Chat view with a rich composer and mentions, next to the terminal
  • YOLO/unattended mode, args and env, set per CLI
  • Multiple accounts per CLI, with plan usage badges
  • Per-project budget limit: warn, pause or kill
  • Fleet: every agent across every project
  • Local Whisper dictation in English, German and Polish
  • Pair a phone on your LAN to watch and steer sessions

Pick the CLI per task, not per vendor.

Teammates

Agents bring in specialists. Questions come to you.

An agent can spawn a designer, a content editor or a developer in its own session, where you can watch it. When any agent needs a decision, you get a form, not a line in a terminal.

  • 4 bundled personas: Universal Dev, Designer, Content Editor, Leader (PM)
  • Fork a persona or add your own
  • Threads: agents message each other directly
  • Agent questions arrive as in-app forms, with a needs-input modal and an attention bar

A question in a terminal scrolls away. A form waits for your answer.

An agent's question shown as a form in ove.run, with answer options and an attention bar.
Skills

The right playbook loads before the first edit.

A skill is a written method for one kind of work: planning a feature, chasing a bug, auditing a UI. ove.run classifies each task and loads the skill that fits. You can also pin a skill to a task or require it in a persona.

Covers process and planning 18
grill-meto-specto-ticketsverify-before-complete
Covers engineering discipline 20
diagnosing-bugssurgical-changestesting-best-practicescode-review-guidelines
Covers specific stacks 5
React, Next.js, React state, React Native, Tailwind
Covers security 3
security-reviewauth-implementation-patterns
Covers design and UI 22
ui-auditvisual-verifyaccessibility-compliancedesign-system-memo
Covers writing 3
writingclarify-copy

Built-ins are read-only. Duplicate one to make it yours, or write a new one.

ove.run Skills view listing bundled skills, with options to create a new skill or duplicate a built-in.
Project knowledge

Knowledge lives in the repo, not in a prompt.

ove.run keeps a project knowledge tree in docs/knowledge/. Every doc is validated, marked stale when the code it references changes, and served to every agent over MCP.

  • A built-in MCP server per session with 64 tools
  • Code index and analysis: search_preambles, trace_symbol, dead code, impact, git_why, LSP
  • Security and type-defect checks
  • Understand: code map, architecture, stack, key files
  • Inspect: bundled analyzers for security, dead code and duplication

The next agent starts from what the last one wrote down.

ove.run knowledge tree with a project doc open and a stale marker on a doc whose code changed.
MCP servers

Configure a server once. Every agent gets it.

Add MCP servers globally or per project. When a session starts, ove.run writes the server list into each of the 8 CLIs' own config files, in that CLI's format.

  • Local command (stdio) or remote URL
  • Add and edit in a form, or Edit JSON in the standard mcpServers shape
  • Chrome DevTools as a one-click preset
  • Turn any global server off for a single project; a project server overrides a global one with the same name
  • Env values are injected at spawn; config files hold only ${NAME} placeholders
  • Existing config entries are merged, and the files ove.run writes are gitignored

One server list. Eight config formats you never edit by hand.

ove.run MCP editor showing global and project-level servers, with a Chrome DevTools preset and an Edit JSON option.
Quality gates

Done means the gates are green.

Set build, lint, typecheck and test commands per project, plus a preview URL. They run when an agent reports completion.

  • Independent critic: a separate agent judges the work against a severity bar you set, until the bar is met
  • Approval grants, per project
  • Write-only credentials: agents use secrets they can never read

The agent that wrote the code doesn't grade it.

Workspace

Work how you like. Shared folder or isolated worktree.

Each new session edits the shared project folder or gets its own worktree, with a new branch and working copy. In Team mode, the whole team shares one worktree.

  • Setup copies .worktreeinclude files and runs your worktree hooks, such as pnpm install; the agent starts only once setup succeeds
  • New task in worktree: one action creates the task, the worktree and the agent
  • Per worktree: Diff, Merge into the main checkout, Create PR, Delete
  • Automations run in the project root, an existing worktree, or a fresh worktree per run
  • Fleet shows each session's ahead/behind, conflicts, and flags overlap when two sessions edit the same file

Run agents in parallel and see where their edits collide.

ove.run New session dialog offering a shared project folder or an isolated worktree.
Git & review

Review every change before it lands.

Each session gets its own change review. Comment on a line and the comment goes back to the agent.

  • Stage and unstage by hunk or line; commit, amend, undo, sign off
  • Fetch, pull (fast-forward or rebase), push, stash
  • Branches, tags, remotes, commit graph, file history
  • Merge and rebase with conflict resolution
  • Branch compare
  • PR status and creation via gh
  • File tree, previews, and a code editor with LSP IntelliSense
Per-session change review in ove.run, with a line comment addressed to the agent.
Capabilities

Everything else you'd open another app for.

ove.run Without ove.run
Agents 8 agent CLIs side by side in one grid A terminal per agent, constant switching
Questions Agent questions arrive as forms Questions scroll past in a terminal
Teammates Agents spawn specialists in their own sessions You brief every new agent by hand
Skills The right skill loads per task; fork or write your own The same instructions retyped into every prompt
Knowledge Validated knowledge tree in the repo, served over MCP The same context pasted into every session
MCP One server list, written into every CLI's config Eight config files edited by hand
Quality Gates run on completion; an independent critic reviews Failures found after the merge
Workspace Shared folder or isolated worktree per session; overlap flagged Agents overwriting each other's edits
Git Hunk-level staging, per-session review Another tool for every review
Tasks MoSCoW board; import from GitHub Issues or YouTrack Issue text copied into prompts by hand
Notes Grid or canvas, quick capture Notes in a separate app
Automation Cron-scheduled agent runs, shell and webhook hooks Every run started by you
Spend Plan usage badges, per-project budget limits A dashboard per provider
History Searchable transcript history Scrolling back through old terminals

Stop switching terminals.

One window for every agent you run.

Download Buy me a coffee