Continuum

Bring your AI subscriptions.

One app for all your agents.

Continuum is the free, vendor-neutral app for Chat and Code. Claude, Codex, Cursor, Grok, OpenCode, Antigravity, Z.ai, and 100+ providers run side by side on the plans and keys you already pay for, with no Continuum fee and no markup. Want Continuum-hosted inference later? Plus starts at $25/mo.

Download for Mac Open the web app

Claude Codex Cursor Grok OpenCode Antigravity Z.ai OpenRouter · 100+ models

ChatUsageCodeSettings

--:--:--

Claude fable-5

01 · Stack your subscriptions

Usage limits stop being limits.

Stack two or three Claude or Codex subscriptions side by side, config-isolated, with a gauge each. When one account's window runs hot, switch to another and keep going. As much compute as you own, zero overage bills.

Toggle between personal and work accounts per session, right from the titlebar. The session never breaks stride when you move it to a fresh window.

Multiple subscriptions, go limitless →

Usage

work personal

work account · 87% used this week

02 · Bring your own subscription or API

ChatUsageCodeSettings

Providers

Claude connected

Codex not connected Connect

OpenRouter

models available via openrouter 0+

2 of 8 providers connectedchanges save instantly

Set up and use your own accounts.

Connect the flat-rate subscriptions and the API keys you already have. Each provider runs through its own official CLI, under its own login - the work lands on your plan, not a marked-up key. No proxy in the token path. No metered anything.

Want a model you don't subscribe to? Bring one OpenRouter key and get 100+ models alongside the rest.

Claude Codex Cursor Grok OpenCode Antigravity Z.ai OpenRouter · BYOK

03 · Broadcast

One message. Every model at once.

Send the same message to Claude, Codex, and Grok at the same time and watch the answers stream side by side, each with its live cost per provider. No more asking the same thing three times in three tabs.

Star the answer that wins and the thread continues with it. The losing runs step aside; you keep going with the model that got it right.

solobroadcast ❯

Claude fable-5 ★

$0.000 tok

Codex gpt-5.5 ★

$0.000 tok

Grok grok-4.5 ★

$0.000 tok

continuing with Claude

04 · Start anywhere, continue anywhere

Start anywhere. Continue anywhere.

Mac, web, and iPhone are peers, not remotes. Start a session from your phone on the train, it shows up on your Mac and in the browser instantly - same session, same branch, same transcript. Pick it up on whichever screen is in front of you.

The session label never changes as you move between surfaces. One running agent, three first-class front ends.

Sessions from anywhere →

iPhone Mac Web

9:41𝗅𝗍𝖾 ▮▮▮

‹ Reykjavíkrk-2

Add retry with backoff to the relay client.

Wrapping the send loop in exponential backoff - 1s to 30s + jitter. Applying the patch.

started here · streaming

05 · Run on any device

Settings

Devices

this mac apple silicon · host live

＋ Add device

1 device · 0 remotelocal-first by default

Close your laptop. The work keeps going.

Sessions don't have to live on the device in front of you. Run them on your own hardware, a cloud server, or serverless - the 2015 MacBook in your drawer, an AWS EC2 instance, a Hetzner box, or Cloud Burst on the Cloudflare and Vercel accounts you connect.

Enroll a device once and the work runs there. Shut your laptop and the agents keep building - all completely under your control, one fleet, one surface.

Run on any device →

06 · Know what you burn

Know exactly what your agents burn.

Spend and tokens broken down by repo, by provider, by model, by day. It's the receipt no metered dashboard gives you.

Spend over time

Claude Codex Cursor OpenRouter Antigravity Grok

24h7d30dAll

$15 $30 $45

Spend by repo

menu bar 63% 41% Wed 21:36

07 · For your whole team

A multi-vendor agent environment for your team.

Bring your staff into one app. Org-wide analytics across every vendor, every member. Subscriptions instead of metered APIs - the same work, at a fraction of the API bill.

Learn More →

team · this month 4 members · $612 spend

Ada L. $214

Grace H. $156

Alan T. $132

Edsger D. $110

08 · Pricing

The app is free.

Inference is optional.

Download Continuum and bring your own provider keys - free app. When you want Continuum-hosted models, personal API keys, and a weekly allowance, pick Plus ($25), Max 100 ($100), Max 200 ($200), or Ultra ($500).

See Free, Plus, Max, Ultra →

metered api · same workload$0.00 and climbing

every token billed · scales with your ambition

your keys + free Continuum$0 Continuum · providers direct

optional hosted plans from $25/mo when you want them

09 · Continuum vs

How we stack up.

Honest matchups against agent workbenches, AI IDEs, and the CLIs Continuum actually runs. We say when the other tool wins.

All comparisons →

vs ConductorParallel agents + worktrees - with mobile and a cost ledger vs CursorKeep the editor. Continuum runs the agent fleet vs Claude CodeClaude Code is the engine. Continuum is the workbench vs CodexCodex beside Claude - one workbench vs WindsurfAI IDE cascades vs multi-device agents vs CopilotCompletions vs long-running agents vs DevinCloud SWE hire vs local-first fleet vs AiderSharp CLI pair vs full product surface vs OpenCodeProvider we run - plus every other lab vs Gemini CLIAntigravity beside Claude and Codex vs HermesSelf-improving agent vs coding control plane vs OpenClawChat-app personal agent vs engineering fleet vs PiMinimal harness vs full multi-device product vs Factory AISDLC Droids vs the plans you own

10 · Guides

How to run these tools.

Pricing you can act on, rate limits explained properly, and the git worktree mechanics behind running several agents at once. Dated, so you know how fresh each one is.

All guides →

Claude Code pricingPro vs Max vs API, with the break-even math Rate limitsThe 5-hour window and weekly caps, explained Git worktreesOne repo, many working directories Parallel agentsSeveral sessions at once without collisions Codex CLISandbox modes, config.toml, and exec Cursor vs Claude CodeWhich one for which job Codex vs Claude CodeTwo terminal agents, opposite philosophies Track your spendTurn local session history into dollars Install Claude CodemacOS, Windows, WSL, and Linux Slash commandsThe built-ins, plus writing your own

11 · Questions

The honest fine print.

Short answers here, the long ones in the docs.

What does it cost?

The Continuum app is free. Bring your own Claude, Codex, or other provider keys and you pay those labs directly - no Continuum fee and no markup. Optional hosted inference is Plus at $25/mo, Max 100 at $100/mo, Max 200 at $200/mo, and Ultra at $500/mo, each step up raising the weekly hosted allowance. See pricing for the full ladder.

How does it use my subscription instead of an API key?

Continuum drives each provider's official CLI under your existing login - in a real PTY or the vendor's own harness, the same way you'd run it in a terminal. Usage lands on your flat-rate plan. Nothing is proxied, marked up, or man-in-the-middled. If you want a model you don't subscribe to, bring an API key through OpenRouter - one key, 100+ models.

Is it really vendor-neutral?

Yes. Claude, Codex, Cursor, Grok, OpenCode, Antigravity, z.ai, and 100+ models via OpenRouter all run in the same Chat and Code workbench. Pick a different agent per session and run them side by side - no vendor is privileged, none is locked in.

Can I really stack two Claude accounts?

Yes. Work and personal Claude (or Codex) subscriptions sit side by side, config-isolated, each with its own gauge. Pin a session to whichever one you like; when one runs hot, switch and keep working. As much compute as you own, no overage bills.

Do sessions keep running when I close my laptop?

Yes, if the session is running on another device. Enroll any Linux or Mac box - an old laptop, an AWS EC2 instance, a Hetzner server - or use serverless Cloud Burst on Cloudflare or Vercel accounts you connect. The work runs there, so you can shut the machine in front of you and check back from any surface.

Where does my code and data live?

On your machines. Transcripts, analytics, and keys are parsed and stored locally on the Mac. Telemetry isn't one of the categories of data that can leave it, and remote traffic is end-to-end encrypted - the relay sees envelope lengths, not contents.

11 · Every platform

Bring your agents to every device.

Native apps for Mac, Windows, and Linux, a terminal CLI, plus iPhone, Apple Watch, and the web - all on one wire protocol. Start a session on one screen and pick it up on any other.

Mac Beta Windows Beta Linux Beta CLI iPhone Watch Web

See all clients and downloads →

12 · Begin

Every agent.

One app.

Bring the subscriptions you already pay for. Chat and Code, every vendor, side by side - free to download, with optional hosted inference when you want it.

free app · your subscriptions · hosted inference optional