Open Source · GPL v3

Your control plane for AI coding agents.

Run Claude, Gemini, Copilot, and Codex side by side. Orchestrate multi-step workflows. Stay in control without babysitting.

The problem

Stop tab-switching.

Terminal Agent Git Notes Issues
Agents

Run any agent.
Run them all at once.

Full support for Claude Code, Gemini CLI, GitHub Copilot, and Codex CLI — all in one interface.

Claude Code Gemini CLI GitHub Copilot Codex CLI
  • Split-screen layouts with drag-to-resize panes
  • Live status — idle, working, needs input, finished, error
  • Unattended mode — agent runs without waiting for your approval
  • Session persistence across restarts
  • Per-agent config — CLI flags, env vars, model selection

Other tools lock you into one provider. ove.run treats agents as interchangeable. Run Claude on your backend refactor and Gemini on your frontend migration at the same time.

Loop Engine

Describe what you want built.
Get working code.

The Loop Engine breaks your request into tasks, assigns them to agents, runs quality gates, and decides whether to proceed, retry, or pause — autonomously.

1
Plan

Request broken into independent tasks with priorities and dependencies.

2
Execute

Agent session started per task with full context loaded.

3
Verify

Quality gates run automatically: build, lint, typecheck, tests.

4
Judge

Arbiter checks results against your requirements.

5
Iterate

Failed tasks retry with adjusted context. Automatic safeguards prevent runaway retries.

on failure, retries from Execute

Most AI tools run one prompt at a time. The Loop Engine runs a full development cycle — plan, execute, verify, judge — autonomously.

Arbiter

Configurable autonomy. Not a binary switch.

The Arbiter monitors agent sessions, reads questions in context, and responds based on a trust level you control.

1 Supervised

Approves each step with you

2 Autonomous

Runs freely, asks when stuck

3 Full Auto

Handles everything

When an agent needs input, a feedback modal surfaces the terminal output, parses options into clickable buttons, and starts a configurable countdown. If you don't respond, Arbiter answers. If you do, it steps aside.

Every decision is logged with full reasoning. Audit exactly why it approved, rejected, or retried any task.

Most tools are either fully autonomous (scary) or fully manual (tedious). The Arbiter gives you a dial. Turn it up as trust grows. Turn it down for sensitive work.

Context

Structured knowledge.
Not a system prompt box.

Context units are reusable blocks — each stored at three levels of detail and scoped globally, per project, or per session. Agents get exactly what they need.

Personas

Role and expertise profiles. Tell agents who they are and how to think.

Skills

Workflow instructions and patterns. Git conventions, test strategies, code style.

Knowledge

Project specifications, architecture docs, and domain references.

References

URLs, files, and external sources scoped to session, project, or global.

Detail levels

L0
Summary One sentence
L1
Overview Structured paragraph
L2
Content Full markdown

Write the full content. Summary and overview are auto-generated — one click, no manual summarization.

12 built-in personas. 8 skill packs covering git workflow, testing, API design, security, and more — ship out of the box.

Capabilities

Everything else you'd open another app for.

ove.run Without ove.run
Agents Claude, Gemini, Copilot, Codex — one interface, side by side Open a separate terminal per agent, context-switch constantly
Orchestration Autonomous loop: plan → execute → verify → retry Manually prompt, check output, re-prompt, repeat
Context Memory persists decisions and patterns across sessions Paste the same context into every new session
Git Branch, diff, stage, commit without leaving Context-switch to terminal or browser tab
Issues Browse and delegate Jira / GitHub Issues in one click Copy-paste issue details into agent prompts by hand
Notes Notes update agent context automatically Keep notes elsewhere, summarize manually per session
Quality gates Build, lint, test run automatically after each task Run tests manually, discover failures after the fact
Oversight Arbiter reviews results, pauses on ambiguity Either babysit every prompt or trust blindly and hope

You've been tab-switching long enough.

Built for developers who delegate.

Download GitHub Buy me a coffee