> ## Documentation Index > Fetch the complete documentation index at: https://continuum-three-olive.vercel.app/docs/llms.txt > Use this file to discover all available pages before exploring further. # FAQ > Frequently asked questions about Continuum: what the free app covers on every platform, what you pay your providers, and where hosted inference fits in. The app is free on every platform — macOS, Windows, Linux, the `continuum` CLI, iPhone, Apple Watch, Android, and the web app — with no feature paywall, when you bring your own provider accounts. You pay Claude, Codex, Cursor, and the rest exactly what you would pay running those CLIs yourself, and Continuum takes no cut. There is exactly one thing you can buy from Continuum: **hosted inference**, a subscription that supplies the models instead of you connecting your own. Plus is $25/month, Max is $100 or $200, Ultra is $500. See [Hosted Inference & Billing](/docs/features/hosted-inference). Accounts exist and are optional. Signing in syncs your devices and manages a plan; skipping it keeps everything local. [Privacy](/docs/privacy) spells out exactly what changes when you sign in. **Bring your own keys (BYOK)** means Continuum drives the provider CLIs under your own logins. Your prompts go from your machine straight to Anthropic, OpenAI, Google, xAI, or Anysphere. Continuum is never in the token path and never sees the traffic. This is free. **Hosted inference** means Continuum supplies the model. Your prompts go to Continuum's gateway, which forwards them upstream using credentials that live server-side and never touch your machine. You get a weekly budget instead of per-vendor subscriptions, personal API keys, and zero setup. This is the paid plan. They coexist. On a paid plan you can still run any session through your own accounts, and many people do — hosted inference for convenience, BYOK for the subscription they already pay for. No. That was true early on and is not true now. * **macOS** is the most mature host, and the reference implementation. * **Windows and Linux desktop** apps (beta) are full hosts too — they run sessions locally, backed by a bundled Go agent. * **The `continuum` CLI** (beta) is both a host and a client: `curl -fsSL https://continuumcode.ai/install-cli.sh | sh`. * **iPhone, Apple Watch, Android, and the web app** are remotes. They drive sessions running on a host; they don't run agents themselves. * A **Linux box, VPS, your own cloud account, or a Continuum-managed runner** can be the host, with no desktop machine involved at all. Sessions always run on a host, because coding agents need real access to files, git, and a terminal. But that host does not have to be a Mac, and it does not have to be the machine in front of you. See [Remote devices](/docs/features/remote-devices). On BYOK, Continuum drives the official provider CLIs under your existing logins and makes no API calls on your behalf. Usage bills through your existing accounts exactly as if you ran those CLIs yourself. OpenRouter and Z.ai via OpenCode are the bring-your-own-API-key paths — you configure the key in OpenCode and Continuum just runs `opencode serve` locally. On a hosted plan, usage bills against your weekly budget instead, and past that against a prepaid balance at pass-through cost with no markup. Yes, and it is no longer limited to Claude and Codex. Nine provider kinds support multiple side-by-side accounts, each with its own isolated config root and credential storage, and sessions pin to a specific account. Pins fail closed: if an account is removed or its token is missing, the session refuses to spawn with a re-authenticate prompt rather than silently billing your default account. See [Multiple accounts](/docs/features/multi-account) for the full flow and the per-provider details. Yes. In **Settings → Devices**, **Add a Linux device** gives you a one-line enroll command — no Tailscale required. Under **Advanced** you can also add a Mac or VPS over Tailscale, connect over SSH, or attach your own AWS cloud. Remote sessions are fully interactive: you steer them from any client, and the running host stays authoritative for the session. See [Remote devices](/docs/features/remote-devices). Partly — and the honest answer is narrower than it used to be. **The agent work is local.** Sessions run on your host against local files, git, and terminals. Continuum does not need to reach anything for a session to run; the provider CLI needs its own network access to the model, and that's it. **Connecting two devices does need the cloud, once.** Since pairing moved to email sign-in, a phone or browser reaching your host mints its connection through Continuum Cloud, which authenticates the account, confirms you own both devices, and sets up the relay channel. There is no LAN-only or fully offline device-to-device mode today. The traffic itself stays end-to-end encrypted and the relay is a blind pipe — but the control plane is not optional. **Tailscale is still supported** as a direct transport between a host and a client on your own tailnet, which keeps session traffic off the relay entirely. It doesn't remove the sign-in step. If you only ever use one machine, none of this applies: a single host with its local UI needs no account at all. Your account identity, which devices are enrolled, your subscription state, and — this is the part people don't expect — a **rolling 30-day mirror of transcripts from recently active sessions**, so a phone or browser can show your work before a host is reachable. Those transcript pages are sealed at rest with a per-member AES-256-GCM key, and removing a member destroys that key so the content becomes permanently unrecoverable. But it is encryption at rest, not end-to-end: Continuum holds the master key, because it unwraps it to render your own history back to you. Signed out, none of this exists — no account, no mirror. [Privacy](/docs/privacy) breaks it down egress by egress, including the internal prompt-inspection path and the append-only audit trail that constrains it, and [the account deletion page](https://continuumcode.ai/delete-account/) covers erasure. Only if you publish one, and only if they have the link. Publishing to `continuumcode.ai/share/` is always an explicit action on a specific transcript. The link is unlisted rather than passworded — its secrecy rests on the id being unguessable — so treat it as "anyone with the URL." You can set an expiry, revoke it at any time, and see a view count. Before it's stored, the transcript is redacted **on the server**, so no client can widen what a link exposes. The redactor rebuilds each message from a known set of fields and drops anything else, masks known secret shapes and home paths and private IPs, and never stores raw tool inputs at all — which is where file contents and full shell commands would otherwise live. Preview exactly what a viewer sees before you publish. See [Sharing](/docs/features/sharing). No. Frames are sealed with XChaCha20-Poly1305 using a symmetric key derived inside your devices via X25519 ECDH. The relay forwards opaque bytes and never holds the key. An operator with full infrastructure access cannot decrypt captured traffic. The relay sees encrypted bytes, a short header (protocol version, sender role, envelope type), byte counts, and your IP. It does not see content. The cloud backend is a separate question with a different answer — see the sign-in question above and [Privacy](/docs/privacy). See [Security](/docs/security) for the full specification. The analytics layer is a Swift implementation of the same ideas behind `ccusage` by ryoppippi: parse local agent JSONL files, normalize token events, deduplicate by `messageId:requestId`, price with a LiteLLM snapshot, and roll up by day and repo. The `ccusage daily` command is the parity benchmark. If Continuum's numbers diverge from `ccusage`, ccusage is treated as the ground truth. The pricing snapshot is bundled with the app and the shipped app never fetches pricing at runtime, so pricing is only as current as the build. A very recently launched model may show as unpriced until a newer build ships. Support, feedback, and feature requests go through [continuumcode.ai](https://continuumcode.ai). There is no public issue tracker to file against. Account deletion requests go to `accounts@montaukanalytics.xyz` from your registered address — see [the deletion page](https://continuumcode.ai/delete-account/). To report a security issue, email the maintainers directly rather than posting it publicly. Continuum is published by Montauk Analytics Inc.