What you get
- Never stop at a usage limit. When one agent runs out, press Enter and the next one resumes the same task exactly where it stopped: same plan, the step that was in flight, and the real code changes so far.
- One shared task across every agent. Every step each agent takes is recorded, so a switch is a resume, not a summary. An agent you return to is caught up on what the others did.
- Only the subscriptions you already pay for. No API keys, no per-token bills, no new account.
- Every agent asks before acting. Edits and commands wait for your approval, just like in each agent's own app, and an edit shows you its diff first.
- Drop files and screenshots into the prompt. Drag an image in and the agent sees it (
[Image #1]); drop any other file and its contents go with your message.
- Undo any turn, or rewind further.
/undo rolls back the last reply's changes; Esc Esc (or /rewind) puts the files back to before any earlier message, and with Claude Code or Codex cuts its conversation back too.
- Checks after every edit. Pick your test or typecheck command in
/config: it runs after each turn that changes files, and a failure can go straight back to the agent to fix.
- See what you're getting.
/usage shows tokens and their API-equivalent value per agent, with daily budgets if you want them.
- Free and open source. Everything stays on your machine.
❯ fix the failing auth test
⏺ Fixed the off-by-one in token expiry. 14 tests pass.
⎿ Claude Code reached its daily limit · Enter to continue in Codex
───────────────────── Codex · gpt-5.6-terra · new session ─────────────────────
❯ now add a test for the refresh path
⏺ Added refresh-expiry coverage. 15 tests pass.
Get started
Any system with Node.js 18+ (brings its own Bun):
npm install -g codemax-cli
Or without npm:
macOS and Linux:
curl -fsSL https://codemax.saidev.codes/install | bash
Windows (PowerShell; Windows support is new):
irm https://codemax.saidev.codes/install.ps1 | iex
It installs codemax (and Bun if you don't have it), then lists the agent CLIs it found. You need at least one agent you're logged in to. Then:
codemax doctor # confirms nothing will bill an API key
cd ~/your-project && codemax
It installs the latest release. Run the same command again to update. To uninstall, delete ~/.codemax/cli and codemax (or codemax.cmd on Windows) from ~/.bun/bin.
git clone https://github.com/SkidGod4444/codemax-cli.git
cd codemax-cli && bun install && bun link
Supported agents
| Agent | Plan it uses | Status |
|---|
| Claude Code | Claude Pro / Max | Supported |
| Codex | ChatGPT Plus / Pro | Supported |
| OpenCode | Your OpenCode login | Supported |
| Grok | SuperGrok | Supported |
| Cursor Agent | Cursor plans | Supported (needs Cursor Agent 2026.10 or newer). Changing its model with /model also changes Cursor's own default. |
| Antigravity CLI | Google AI plans | Supported. It has no ACP mode of its own, so codemax offers to download Google's official ACP server for it (about 110 MB, once, from dl.google.com) the first time you switch to it, and signs in with your Google account (a browser may open once). It reports context size but not token counts, so /usage shows input only. |
| Copilot CLI | GitHub Copilot plans | In its own UI via codemax --tui |
Everyday keys
| |
|---|
| ↑ on an empty prompt | Your earlier messages in this project, kept across restarts |
| Ctrl+R | Search your earlier messages; Enter puts one in the prompt |
| ↑ with messages queued | Edit a queued message (↑ again for the one before); Enter puts it back, clear it and Enter to remove it, Esc leaves it as it was |
/switch or Ctrl+^ | Move to another agent, conversation included |
/ask codex <msg> | One question to another agent |
@file | Attach a file to your message |
/task | The shared task: plan, who did what, where it stopped, files changed |
/debate <goal> | 2-3 agents, on models you pick, each propose a plan, argue it out, and a judge merges the best; then build it with your agent |
Ctrl+G or /editor | Write your message in your own editor |
/copy, /export | Copy the last reply; save the conversation as Markdown |
/setup | Go through every setting one by one (shown once on first start) |
/update | Update codemax when a new version is out, then restart into it with the same agent and session (codemax update updates without starting) |
/config | Every setting in one panel: theme, keep awake, side agent and its model, handoffs, snapshots… |
| Side agent | Turn it on in /config and pick its agent and model. It reviews the work in the background and speaks up only when something needs attention: its note waits above the prompt, where 1 expands it, 2 sends it to the main agent to discuss, 0 dismisses it. It never uses the main agent's plan unless you allow that in /config |
/awake on, /awake lid | Keep the machine from sleeping while agents work; lid also with the lid closed |
/undo | Revert the last turn's changes |
~/.codemax/config.json is created on first run. codemax config prints it.
| Key | Default | What it does |
|---|
order | ["claude", "codex", "opencode", "cursor", "grok", "antigravity", "copilot"] | Which agent comes next on a handoff |
budgets | {} | Daily limit per agent, e.g. {"claude": 20, "all": 50} |
routes | [] | Send matching messages to a specific agent |
checkpoints | true | Snapshots for /undo and /rewind |
verify | {} | Checks per project, run after each turn that changes files, e.g. {"/path/to/repo": "bun run test"} (set it in /config or with /verify <command>) |
verifyOnFail | "show" | "show" attaches a failure to your next message; "fix" sends it to the agent (at most twice per task) |
autoSwitch | true | Offer a handoff when an agent runs out |
summarizeOnSwitch | true | Ask the outgoing agent for a summary before a switch. Only used for history recorded before the step journal; otherwise the next agent resumes from the recorded steps |
priceUpdates | true | Weekly price list refresh |
hotkeys | ["ctrl+/", "ctrl+^"] | Keys that open the agent switcher |
codemax [--agent <id>] [--fresh] start chatting: continues this folder's latest thread (--fresh: a new one)
codemax --resume [<id>] reopen a thread (no id: pick one); add --agent <id> to continue it with another agent
codemax threads this folder's threads, newest first
codemax doctor [--fix] check nothing bills an API key
codemax agents installed agents and their quota
codemax usage [--all] usage for this project or all projects
codemax prices [update] price list used for usage values
codemax history | brief | status | config | keys
- Your code goes only to the agents you already use, under your own logins.
- codemax always removes vendor API keys from the agents it starts, and
codemax doctor --fix removes any left in Claude Code's settings.
- Grok follows your Claude Code permission setting. If that is
auto, Grok acts without asking, and codemax warns you.
- Its own network calls are a weekly price-list download (
priceUpdates: false turns it off) and a version check at most every 12 hours against npm or GitHub releases (updateCheck: false turns it off). Both are in /config.
- Usage values are what your plans covered at public API rates, not money you spent.
Contributing
Issues and pull requests are welcome. First-time contributors sign a short CLA on their pull request. See CONTRIBUTING.md and PLAN.md for design notes.
bun test && bun run typecheck
Not affiliated with Anthropic, OpenAI, xAI, Google or the OpenCode team.
License
Copyright © 2026 Saidev Dhal. Free and open source under the GNU AGPL-3.0, with an added permission for the agent tools codemax drives.
You can use, change and share codemax. If you distribute a modified version, or run one as a service, you must publish your changes under the same license. The codemax name and logo are not covered by the license: forks need their own name. See TRADEMARKS.md.
