uv tool install tandem-cli

Quick start

You'll need Python 3.11+ and the claude and codex CLIs on your PATH.

uv tool install tandem-cli   # or: pip install tandem-cli
cd your-project
tandem                       # fresh paired session; drops you into claude

Work normally — that's the real Claude Code TUI. When you want the other model, press Ctrl-]: tandem closes out the harness at the turn boundary and reopens the same conversation in Codex a couple of seconds later. Press it again to come back. The bottom row tracks who you're facing:

claude ● │ codex ○   ^] flips

Pressed mid-turn, the flip arms and fires the moment the model finishes (the bar says so; press again to cancel).

Exit the harness the usual way and you're back at your shell, with the session saved and a hint for picking it up again:

to continue this session: tandem resume a1b2c3d4e5f6

Come back anytime:

tandem resume                # most recent session in this directory
tandem resume a1b2c3d4e5f6   # a specific one (id from the exit hint)

Why tandem?

🖥️ One CLI, two harnesses, zero ceremony.

tandem is the terminal you live in. It fronts the real Claude Code or Codex TUI — pixel-for-pixel native — and Ctrl-] flips to the other one in a couple of seconds, same conversation, same files, same history. A one-line tab bar on the bottom row shows which model you're facing; everything above it is the untouched native UI.

🆕 0.2 — GPT subagents inside Claude Code

Type "get the code reviewed by gpt" and tandem's plugin hands the task to a local codex worker; the verdict lands back in your Claude session, and your Claude quota stays on the main thread. Pin any model your codex account offers ("ask sol to review it") or set a cheap default once. Setup and routing: GPT subagents guide.

⏳ Hit a usage limit? Just keep going.

Claude runs out of its usage window mid-refactor? Press Ctrl-] and Codex continues the same conversation a second later — same files, same history, same plan. Two subscriptions become one long runway.

🌩️ Immune to outages.

An Anthropic or OpenAI outage doesn't stop your work: Ctrl-], keep going in the other harness, and flip back whenever it clears.

Five more reasons — subscriptions not API bills, two model families on one problem, native harnesses, instant switching, privacy: Why tandem?

How it works

• Only one model runs per turn — the other side is never invoked to "catch up".
• As you work, tandem translates the growing transcript into the other CLI's native session format — pure local file I/O, no model calls — so the other side is always resume-ready.
• Each CLI is the real thing running on a PTY: your keybindings, slash commands, and MCP servers all work exactly as they do today — tandem reserves exactly one key (Ctrl-]) and one terminal row (the tab bar).
• A flip waits for the turn boundary, exits the fronted CLI gracefully, lets the sync settle, and resumes the other side — no stop in between.
• Everything stays local: the CLIs' own session files plus a small SQLite database in ~/.tandem. No cloud sync, no telemetry.
• Uninstall tandem tomorrow and both sessions still resume natively with claude --resume and codex resume.

Full mechanics — sync engine, crash safety, compatibility ranges: How tandem works.

Command cheat sheet

Three maintenance commands round it out: tandem doctor (health check), tandem sync (manual catch-up translation), and tandem sync-mcp (share MCP server configs between the tools).

Docs

• GPT subagents — plugin install, worker model, routing modes, sandbox & trust boundary
• Why tandem? — the full pitch, all eight reasons
• Configuration — the optional ~/.tandem/config.toml: subagent workers, per-harness startup args, the flip key and tab bar
• How tandem works — sync engine, PTY passthrough, compatibility, where your data lives
• Developing tandem — dev setup and the converter adapter interface
• Observed session formats — the claude/codex transcript details tandem is pinned against

License

MIT