> ## 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.
# Hosted Inference & Billing
> Continuum is free with your own provider accounts. Hosted inference is the optional paid add-on: what the tiers buy, and how budget and wallet work.
Continuum has exactly one thing you can pay for, and it is not the app.
The app is **free on every client** — macOS, Windows, Linux, the `continuum` CLI, iPhone, Apple Watch, Android, and the web app — for as long as you bring your own provider accounts. **Continuum-hosted inference** is the optional subscription: instead of connecting your own Claude, Codex, or Cursor logins, you buy inference from Continuum with a weekly budget and a single set of credentials.
Bring-your-own-key (BYOK) and hosted inference are additive, not exclusive. On a paid plan you can still run any session through your own provider accounts, and Continuum never sits in that token path — you pay Anthropic, OpenAI, and the rest exactly what you already pay them.
## The two ways to run models
| | Bring your own accounts | Continuum-hosted inference |
| --------------------------- | --------------------------------------- | ---------------------------------------- |
| What you pay Continuum | \$0 | $25–$500 / month |
| Who bills the tokens | Your provider, directly | Continuum, against a weekly budget |
| Setup | Install and log in to each provider CLI | Sign in once; nothing else |
| Credentials on your machine | Yours, in your Keychain | None — the upstream keys are server-side |
| Available on | Every client | Every client |
## Plans
Prices and weekly budgets as published on [continuumcode.ai/pricing](https://continuumcode.ai/pricing/):
| Plan | Per month | Weekly hosted budget |
| ------- | --------- | -------------------- |
| Free | \$0 | Free-model lane only |
| Plus | \$25 | \$25 |
| Max 100 | \$100 | \$100 |
| Max 200 | \$200 | \$200 |
| Ultra | \$500 | \$1,000 |
Every plan — including Free — keeps the whole app, every provider via your own keys, and local usage analytics. What the paid tiers add is Continuum-routed frontier models, a weekly spend budget, and [personal API keys](#personal-api-keys).
### What "weekly budget" actually means
The window is **seven days from when your plan started**, not a calendar week. It rolls forward from that same instant — a plan that began Wednesday at 09:14 resets every Wednesday at 09:14. This is deliberately the same absolute-reset shape the provider CLIs use, so a hosted budget and a Claude or Codex weekly limit behave the same way.
When the weekly budget runs out, what happens depends on your plan:
* **On a paid plan**, further usage draws from your prepaid **usage balance** at pass-through model cost with no markup. If that balance is empty too, requests stop with *"Weekly hosted-inference budget reached. Add prepaid funds or upgrade your plan."*
* **On the Free plan**, the free lane simply stops until the window rolls: *"You've used this week's free hosted chat. Upgrade for a larger weekly allowance."*
* **In an organization**, an admin can cap you below your plan and decide whether you may exceed it. If they capped you with overage off, you'll see *"You've reached the hosted-inference allocation your organization set for you. Ask your admin to raise it or enable extra usage."* See [Organization & Cloud](/docs/features/organization).
### Free hosted models
Every plan, including Free, can reach a set of free hosted models — and signed-out visitors get one free message per week before signing up.
The roster is set by the server, not by the app, so it changes without a client release. **Look at the model tray for the live list** rather than trusting any written enumeration, including this one. Free models are flagged as such in the tray and grouped under **Included in Free**, with the paid models under **Unlock frontier models**.
## Turning it on
Hosted inference appears as a provider named **Continuum**, alongside Claude, Codex, Cursor, and the rest.
Open **Settings → Account** and sign in with your email. This is the same account that syncs your devices; see [Remote devices](/docs/features/remote-devices).
In **Settings → Account**, the **Inference plan** card shows your current status. **Get hosted inference** opens Stripe Checkout in your browser; **Manage billing** opens the Stripe portal to change tier, update a card, or cancel. On iPhone the purchase runs through the App Store instead, and the card reads "Managed through the App Store on your iPhone."
In **Settings → Providers**, connect your Continuum account. Continuum mints a personal key and configures the local OpenCode runtime for you — there is no key to copy, paste, or store. Disconnecting revokes that key server-side and strips the configuration back out.
Hosted Claude, GPT-5.6, Grok, and Gemini models appear in the model tray under **Continuum**. Sessions you start on them bill against the weekly budget instead of any provider subscription.
If you pick a model your plan doesn't include, the session refuses to start rather than silently downgrading: *"That model needs Continuum hosted inference. Upgrade to Plus, Max, or Ultra — or connect your own provider key."*
## The usage balance (prepaid overage)
The **Usage tab** carries a **Usage balance** card — a prepaid wallet that covers usage past your weekly plan budget, drawn at pass-through model cost with no markup.
* **Add funds** takes $5 to $10,000 and opens Stripe Checkout.
* **Auto-top-up** adds a fixed amount whenever the balance drops below a threshold you set. It needs a card on file; the card warns you if there isn't one yet.
* **Recent activity** is the ledger of top-ups and draws.
The balance is only consulted on paid plans. Free accounts get the upgrade nudge instead of an overage path.
## Personal API keys
Paid plans can mint personal keys that let *other* tools spend your hosted budget — opencode, Claude Code, Cursor, or anything that speaks the OpenAI or Anthropic HTTP shape.
Manage them in the **Usage tab** under **Inference API keys**. **Create key** shows the value exactly once ("Copy your key now. It won't be shown again"); after that only a `prefix…last4` stub is listed, with **Revoke** next to it. Keys start with `cont_sk_`.
Both API shapes are served from the same base URL, `https://continuumcode.ai/v1`:
| Shape | Endpoint |
| -------------------- | --------------------------------------------- |
| OpenAI-compatible | `POST /v1/chat/completions`, `GET /v1/models` |
| Anthropic-compatible | `POST /v1/messages` |
Point the usual environment variables at it — `OPENAI_BASE_URL` / `OPENAI_API_KEY`, or `ANTHROPIC_BASE_URL` / `ANTHROPIC_API_KEY`. The panel prints ready-to-paste snippets. Keys are a paid-plan feature: creating one on a free account returns *"Upgrade to a hosted plan to issue an inference API key."*, and the model catalog re-checks entitlement on every call, so a key minted before a downgrade stops seeing paid models.
## What hosted inference means for your data
This is the one lane where prompts reach a Continuum-operated server by definition — that is what you are buying. Be deliberate about it:
* **Your prompts and completions traverse Continuum's gateway** on their way to the upstream model. Sessions run through your own provider accounts do not.
* **Upstream provider credentials are server-side only.** They never reach any client UI, and at rest they are sealed with AES-256-GCM under a key the database never holds.
* **The gateway is the enforcement point**, not just the UI. Organization model allowlists and spend caps are applied server-side, so a modified or older client cannot route around them.
The full picture — including what is retained and for how long — is on the [Privacy](/docs/privacy) page.
## For organizations
An organization admin governs hosted inference for the whole team from **Settings → Your Organisation → Hosted inference**: which models each team or member may use, weekly and monthly spend caps, whether a member may exceed their cap, comped seats, threshold alerts, and an inbox for members requesting more budget.
Those controls are documented in [Organization & Cloud](/docs/features/organization).
## See also
Org-wide inference governance, usage rollups, and the admin surfaces.
The BYOK side: every provider Continuum can drive with your own accounts.
Where the hosted budget, gauges, wallet, and API keys live.
Exactly what leaves your machine on each of the four data postures.