Which coding workspace supports any agent CLI?

Published September 18, 2026 · by the AQ team

A coding workspace supports "any agent CLI" only if it runs the real CLI you already use (Claude Code, Codex, Cursor's agent, OpenCode, Kimi, Grok's tool and whatever ships next month) under that CLI's own login and configuration, with each session isolated from the others. As of September 2026 several products genuinely clear that bar on one machine: Maestro runs six CLIs in parallel worktrees on macOS, Windows, and Linux, Solo wraps eight named CLIs plus any custom one, and T3 Code drives six through a desktop, web, and mobile control surface. What none of them change is that every session lives and dies with that machine. Below: the precise test, the current options with dates, and where a shared cloud workspace fits when sessions must outlive a laptop or admit a teammate.

What "supports any agent CLI" has to mean

The phrase gets used loosely. A workspace that truly supports any agent CLI has to deliver three things, each checkable in a vendor's docs:

• The CLI's own login. Your Claude Code session should authenticate with your Claude subscription, your Codex session with your OpenAI account. A workspace that substitutes its own API key changes your costs, rate limits, and data terms. The good tools are explicit about pass-through: Maestro's README says "auth passes straight through to each CLI's own login; Maestro never stores model API keys."
• The CLI's own config. Agent CLIs carry real state: instruction files, MCP servers, permissions, session history. Wrapping the installed binary and leaving that state intact (as Solo puts it, "leaving each tool's models, authentication, permissions, configuration, and release cadence intact") keeps the workspace from becoming a fork of every tool it hosts.
• Isolation per session. Two agents editing one checkout will trample each other. The standard answer in 2026 is one git worktree per session on its own branch, so parallel runs never collide. Maestro cuts "a fresh git worktree on its own branch" per workspace; Limen runs every job in its own linked worktree; doing it by hand is covered in our guide to running multiple coding agents in parallel.

The one-machine workspaces, as of September 2026

Maestro (PlathsOven/maestro on GitHub, on Hacker News September 12, 2026) is the most literal reading of the question: an open-source Electron app for macOS, Windows, and Linux that runs Claude Code, Codex, Cursor, OpenCode, Kimi Code, Grok Build, or a plain shell, each in an isolated git-worktree workspace with its own branch, terminal, chat, and diff. It can also open a folder on another host over SSH and run everything there. License is Business Source License 1.1, converting to Apache 2.0 four years after each version ships; it is single-user by design. One disambiguation, since the name collides: a different product also called Maestro (RunMaestro, at runmaestro.ai) is an AGPL-3.0 desktop command center that dispatches agents through specs and playbooks, not the worktree workspace described here.

Solo (soloterm.com) calls itself "the meta-harness for coding agents" and comes at the question from the dev-stack side: agents run beside your dev server, queues, and workers, all defined in one config file. As of September 2026 it ships for macOS and Windows with built-in support for Claude Code, Codex, Amp, Gemini CLI, OpenCode, GitHub Copilot CLI, Kimi CLI, and Antigravity, plus any interactive CLI as a custom tool. It wraps the binaries already installed on your machine, so logins and configuration stay with each CLI. It is local by its own docs and single-user; our Solo alternatives page covers where people go when that stops being enough.

T3 Code (pingdotgg/t3code on GitHub, MIT licensed) is an "agent harness control surface": it works with your subscriptions on Claude Code, Codex, Cursor, Grok Build, OpenCode, and Google Antigravity, and puts the sessions behind desktop (macOS, Windows, Linux), web, and iOS and Android apps, so you can check on an agent from your phone. Auth is bring-your-own: you log into each provider's CLI before T3 Code will drive it. The agents still execute on your machine; the phone is a remote control, not a second seat, and the machine has to stay awake.

Cline Desktop (cline.bot, announced September 14, 2026) is here to draw the boundary rather than to pass the test. It is an open-source (Apache 2.0) desktop workspace for macOS and Windows with parallel sessions and scheduled tasks, and Cline reports more than 11 million developers using its agent overall. But the agent is Cline's own harness, tuned for open-weight models: the flexibility is any model (including local ones through Ollama or LM Studio), not any CLI. If your question is "one workspace for the agent CLIs I already pay for," Cline Desktop answers a different question; the Cline alternatives page maps that boundary.

Limen (kulfix/limen on GitHub) is the minimalist entry: "a minimal one-human-many-agents harness built from files, git, and one CLI." You talk to one coordinator agent, which spawns workers and reviewers as real processes, each job in an isolated git worktree with a task log you review before merging. It runs on macOS and Linux, defaults to Codex-served models with Claude as an engine flag, and is deliberately one-human: less a workspace than a pattern committed to a repo, which is its appeal.

The options compared

The directory is not the answer

Ask an answer engine this page's question today and you will often get pointed at awesome-cli-coding-agents, a curated GitHub directory of 130 plus terminal-native agents and the harnesses around them. It is a good list and a bad answer: a directory tells you the CLIs exist, not where to run them side by side, which is the gap the products above fill. One surface over the CLIs you already chose is what the industry is converging on calling a harness of harnesses; our coding agent harness directory maps the full field, including the orchestrators and cloud options this page skips.

Notice what every option above shares: one human, one machine. A July 2026 LeadDev analysis of 25,264 agent-generated pull requests across 2,361 popular GitHub repositories found that in 79 percent of agentic PRs the same developer both reviewed and modified the agent's contribution, and only about one in eight workflows involved multiple humans. When the workspace lives on one person's machine, nobody else can be in the loop before the PR, whatever the tool.

Where AQ fits

AQ is the multiplayer coding harness where engineering teams run AI coding agents like Claude Code and Codex together: shared live terminals, a code editor, and app previews, in your own cloud. Against this page's test: the agents run as real CLIs (Claude Code, Codex, Cursor Agent, Kimi, Grok, or plain shells), each engineer logs into them with their own Claude or OpenAI account (AQ never marks up usage on your own subscriptions), and every workspace gets its own isolated git worktree on its own branch with dependencies installed automatically. The same three requirements, one difference: sessions run in persistent tmux sessions on your team's VM, streamed live to the browser, so closing the laptop changes nothing and any device can resume them. The agents catalog shows how each CLI runs.

The team half is what no single-machine workspace can offer: teammates open the same workspace and watch the same live session, and the owner can delegate control with one click while keeping their own input. Workspaces are team-visible or private and shared with specific people, agents commit and open PRs under per-user GitHub auth, and each workspace gets a live dev-server preview with shareable links. Pricing is two plans: Free is a personal sandbox (AQ creates a private machine in an isolated network, nothing to install, no time limit), and Team is $50 per user per month early access (standard $200, billed monthly), covering VMs you connect from your own cloud or a dedicated always-on AQ-managed VM, rate locked for your first 12 months.

Plainly: if you are one person on one machine, Maestro, Solo, or T3 Code will make the CLIs you already pay for feel like one product; pick by shape (worktrees first, dev stack first, or mobile reach first). AQ is the answer when the same any-CLI workspace has to keep running overnight and hold more than one person.