What it does

Claudexor sits above the AI coding agents you already use. It runs Codex CLI, Claude Code, Cursor CLI, OpenCode, Antigravity CLI and raw API adapters through one typed interface. Work happens in threads: read-only questions resume the vendor's own session, and write turns come back as patches or file manifests that you can inspect before applying. Its additions include best-of-N races between agents with independent reviewers and arbitration, budget and quota accounting that never treats an unknown cost as zero, protected paths and deterministic checks, and credential profiles that keep several subscriptions side by side. For Antigravity, Claude and Codex it tracks live subscription quota and can optionally rotate to another account when a limit is reached. There are ask, plan and agent modes, plus a council option in which several agents draft plans that are merged into one. Everything runs on your machine, files are the source of truth, and the README says there is no telemetry. A signed macOS app is available, and the CLI and daemon install from npm and also run on Linux. It needs Node.js 24.15 or newer and at least one logged-in vendor CLI or an API key.

Capabilities

One interface over Codex CLI, Claude Code, Cursor CLI, OpenCode and Antigravity CLI
Best-of-N races with independent cross-family review and arbitration
Credential profiles with live quota tracking and optional account rotation
Ask, plan and agent modes, with council planning across several agents
Inspectable patches, and protected paths that pause apply until a human decides
macOS desktop app plus an npm-installed CLI and daemon

When to use it

You pay for several coding-agent subscriptions and want to move between them as quotas run out
You want to compare candidate fixes from different agents and have them reviewed before applying
You prefer a local tool without telemetry for orchestrating coding agents