The Continuum app is free. Bring your flat-rate plan or API key from any lab or inference company, without a markup. Continuum-hosted inference is optional when you want it.
Claude
OpenCode
Antigravity
OpenRouter · BYOK
From any company or inference provider. Continuum drives each one through its own CLI and its own login, so the work lands on the plan you already pay for.
Continuum is not another agent. It's the workbench the agents you already run live inside: same binaries, same models, same plans, one place to start them, watch them, and pay for them. Adding a provider takes a login, not a migration, and removing one takes a toggle.
Five of the eight ride a flat-rate plan you already pay for: Claude, Codex, Antigravity, Grok, and the Z.ai GLM Coding Plan. Three are bring-your-own-key: Cursor, OpenRouter, and OpenCode. Nothing forces you to pick a lane. A subscription session and a BYOK session run at the same time, in the same window, against the same repo.
That mix is the point. A flat plan is the cheapest way to run the bulk of your work and the worst way to reach a model it doesn't carry; a metered key is the reverse. Continuum lets you hold both and decide per session, which is a decision you can only make well if you can see what each one costs. The pricing comparison lays the plans out side by side, and Cursor's pricing is the usual first surprise.
Detection is by binary, not config. Continuum looks for claude, codex, cursor-agent, grok, opencode, and agy on disk. A provider goes live when its CLI is present and the provider is switched on in Settings; anything missing shows an install action instead of a dead row. Each provider also has its own on and off switch, so you can keep a CLI installed and still keep it out of the pickers.
Everything you connect works from the Mac app, the web app, and your iPhone. One set of subscriptions and API keys, reachable from wherever you are.
Close your laptop - sessions keep running on any device you've enrolled. Start a run at your desk, check it from your phone, pick it back up in the browser.
Claude · portland
portland
The same enrolled devices, one login each. Laptop closed, the run keeps going.
Three steps, and none of them route a token through Continuum.
Download the DMG. It detects the provider CLIs you already use - claude, codex, cursor-agent, grok, opencode - and lights up a gauge for each one it finds.
Every CLI signs in with its own login - Claude's Keychain OAuth, codex login, Cursor's own auth. Tokens never leave your machine and never pass through Continuum.
The agent runs under your existing plan. Claude Max, ChatGPT / Codex, Cursor Pro, SuperGrok - the work bills the flat rate you already pay, at the vendor.
Where each credential lives is worth knowing, because it's the reason none of them can be rerouted. Claude keeps a CLI-managed credential and its OAuth token in the Keychain. Codex stays CLI-managed through its own app server. Antigravity keeps its state under ~/.gemini. Grok reads its own config root. Cursor takes an API key you paste into Settings. The three OpenCode connectors - OpenCode itself, OpenRouter, and Z.ai - share opencode's own auth.json, one key per vendor.
Nothing here is a Continuum-specific auth scheme. Sign in once with the Claude Code CLI, the Codex CLI, or the Cursor CLI and Continuum uses that login. Sign out in the CLI and Continuum loses it too. The credential belongs to the tool that issued it, which is exactly the property that makes "no markup" verifiable rather than a promise.
No proxy in the token path. No markup. Ever.
A gauge isn't an estimate off your logs. It's the provider's own quota reading, polled live, rendered the same way for every vendor so you can compare six ceilings without learning six dashboards.
Claude, Codex, Antigravity, Cursor, Grok, and Z.ai each report a live window. Continuum draws it as a radial dial with the fraction consumed and a reset countdown beneath. The label follows the vendor: session for Claude and Codex, 5 hour for OpenCode, billing period for Cursor, weekly limit for Grok.
Providers with a second, slower window get a pill bar under the dial. Z.ai reports both a 5-hour and a weekly cap. Cursor splits its monthly rails into Auto and API. OpenCode carries a monthly rail alongside its weekly one, so its ladder runs 5 hours to weekly to monthly.
OpenCode publishes quota only through its web dashboard, so its gauge needs a workspace id and a session cookie. OpenRouter has no quota to report at all and tracks as a dollar-cost lane instead. Both still land in analytics with full token and cost history.
Any enabled provider can put a gauge in your menu bar: percentage, reset countdown, logo inline. Click it for a compact panel with a tab per provider, live percentages, and a stale pill when a reading is cached rather than fresh. Cursor and pay-as-you-go OpenCode default to off; the rest default to on, and every one of them is a checkbox.
Short answers here, long answers in the docs.
No. Gauges and sessions light up per CLI found on your machine. Plenty of people run Continuum with a single Claude Max plan; it gets more useful with every provider you add.
Yes. OpenRouter BYOK sits beside your flat-rate plans - bring a key for any model outside your subscriptions and run a subscription session and a bring-your-own-key session in the same window.
No. Each provider's CLI authenticates itself under your own login; keys stay in your keychain. Continuum never proxies, marks up, or man-in-the-middles provider traffic.
Yes - the Z.ai GLM Coding Plan is a first-class provider: its own login, its own live 5-hour and weekly gauges, and its own column in analytics. It rides opencode's connector path, so its key lives in opencode's own auth file rather than a Continuum-specific store.
The provider shows an install action instead of a dead row, with the same command Continuum's onboarding uses. Nothing is silently unavailable and nothing is silently substituted: a provider is only selectable when it's enabled, authenticated, and reachable. Installing the CLI by hand works exactly the same way - Continuum picks it up on the next scan.
Yes, on every provider. Each account gets its own config root, its own Keychain partition, and a scrubbed environment, and every session is pinned to one before it spawns. See multi-account for how the isolation works and what it refuses to guess at.
Continuum finds your CLIs and lights a gauge for each.
no markup · your login, never ours