Cost & model access
What Foxl credits pay for, where they come from, and why Foxl Agent runs on your own model access
Free accounts can use Foxl's features; managed cloud work still spends credits. Start with the Free tier guide for included features, grants and limits. The model-access path determines how usage is billed, even when products are opened from the same app.
| Path | Model access | Model bill |
|---|---|---|
| Code chat and managed Remote coding agents | Foxl-managed Amazon Bedrock through Relay | Foxl account credits |
| Remote Kiro and Cursor | Your vendor API key saved in Code settings | The vendor bills models; Foxl bills compute only |
| Foxl Agent chat and Notes (on the desktop, and in the phone and web apps through your desktop), and Foxl Code with Where Foxl Code runs set to This computer | Your provider key, local model, or supported subscription | Your provider/subscription |
| Remote chat while your computer is offline, when offered and enabled | Foxl's cloud models for your account's access path | Foxl account credits for model usage and compute |
| Notes AI in the cloud, or Foxl-hosted Agent inference, on accounts enabled for those services | Foxl-managed models | Foxl account credits |
| Notes transcription and voice input | Your selected engine; on Automatic, your own AWS account or the on-device engine first, otherwise Foxl's own Amazon Transcribe | Your provider bills its own key/account; on-device costs no credits; Foxl's own Amazon Transcribe (the Foxl Relay engine) uses Foxl account credits, about 0.4 a minute of audio |
| A phone call's voice | The AWS account set up on your Mac; on a Mac with none, Foxl's own cloud service | Your AWS account; on a Mac with none, Foxl account credits for what the call used, at most about 1.39 a minute |
See Code model access for the managed catalog and AI Providers to connect your Foxl Agent providers.
There is no plan to buy
Foxl sells no plan or credit top-up. The Account page shows no plan, and there is nothing to upgrade to. Free accounts receive the welcome and monthly grants described below.
Credits pay for Foxl Code - on code.foxl.ai and in the app's Foxl Code, which runs on Remote by default on every device: Code chat, managed coding-agent turns, scheduled and review runs, and Remote compute. Free accounts can use every model in the managed Code catalog, including Opus, once admitted to Code's private beta. Credits do not grant beta access by themselves.
Foxl as a model provider is shown and served only to accounts enabled for it, whatever the plan, so an ordinary account, Free included, does not see it. Transcription on Foxl credits is for every account signed in to Foxl: Notes recordings, audio imports and voice input use it where nothing else can transcribe, and always with the Foxl Relay engine picked (Notes credits & privacy). Remote chat and Notes AI in the cloud have their own availability and opt-in, and spend credits when used. Foxl Agent and Notes using your own provider, and Code on This computer, do not debit the Foxl wallet.
Foxl Agent: bring your own access
Under Settings > Model & provider, connect a provider API key, run Google's open Gemma 4 under On this device on a Mac with Apple Silicon, a Windows PC or a Linux x64 computer, use Ollama / vLLM / LM Studio, or sign in with a supported Claude, ChatGPT, or Gemini CLI account. Foxl does not add a model charge for these local paths. Your provider's usage limits, subscription terms, and any extra-usage charges still apply.
The desktop relay tunnel used to reach your computer from another device is separate from managed model inference. A request sent to your Foxl Agent provider stays on that provider's billing path.
Code: managed credits
Code chat, managed coding-agent turns, scheduled orchestration, and reviews use Foxl Relay. Remote compute is metered too. No personal API key, subscription upload, or AWS account is required for managed models. Kiro and Cursor use your vendor API key and vendor model billing instead; Foxl bills compute only for those agent runs. A connected machine supplies its own compute while the models for its managed Code tasks still draw on credits.
One account wallet holds Foxl credits. 1 credit has a displayed value of $0.04 and covers $0.036 of model-provider cost. The meter reserves credits before a request, settles actual reported token usage afterward, and releases the hold. Foxl Agent BYOK/BYOA model calls do not debit it.
When a GPT model in a Foxl Code task (a Codex task, for example) searches the web with Amazon Bedrock Web Search, each search costs $0.012, a third of a credit, on the same usage line as the reply. Opening a page from the results is not charged.
See Budgets and tiers for the current model rates, long-context pricing, compute charges, and the limits of per-task budgets.
Requests at once, and held credits
A Free account runs at most 32 managed model requests at once. A request past that is refused with "Too many requests in flight" until one finishes, and Foxl Code refuses a new managed coding task up front ("no free model slot") rather than starting one that would fail on its first model call; for an account's first 7 days Foxl Code starts at most 4 such tasks at once. Opening a terminal session does not need a free model slot, but the model requests its agent makes count toward the limit. Requests running at once share the balance: one that does not fit beside the credits the others are holding is refused for credits, so on a small balance fewer run together. A coding agent refused that way is asked to retry every 10 seconds while the others finish, and stops once its own retries run out (a few minutes). A Free account also makes at most 200 managed model requests in any 24 hours, Foxl Code's chat and coding agents together (remote chat's cloud replies are not counted; your credits bound those), and is refused with "Rate limit exceeded" past that. See Budgets and tiers.
Each running request holds its reserved credits and its slot. A request that is cut off partway gives both back within about 90 seconds; one cut off after it had already run for several minutes can keep them for about as long as it had run, and never more than 15 minutes. A request that keeps running keeps its slot until it finishes and is charged once, for what it used.
What increases model usage
- Longer conversation history is sent again as input on later turns.
- Reasoning consumes output tokens.
- Tool results and additional agent turns add model requests.
- Concurrent coding agents, scheduled work, and background reviews each use their own requests.
- Cache reads can reduce input cost; cache writes have their own rate where supported.
Credit refresh and the usage window
A Free account receives 10 monthly credits on its signup day-of-month at midnight UTC, starting the month after it signed up. Monthly grants do not accumulate across periods: each one expires when the next arrives. Welcome credits are separate: 200 for Google, Apple, or GitHub signup and 50 for email signup, with a password or a sign-in link, granted once per mailbox and without a monthly expiry. Changing sign-in methods does not grant another welcome allowance. New credits first pay down a negative balance. Any unpaid part carries forward; an amount already repaid is not deducted again at a later refresh.
The period shown in your account follows that grant date, not necessarily the first of the month. An anchor on a missing day in a shorter month uses that month's last day, then returns to the original day in the next month. Your account shows the next refresh date.
Workspaces and shared storage
Every account, a Free one included, can create or join a workspace with team Channels, DMs and huddles. A workspace whose billing owner is on Free is charged nothing for storage or Channels, but accepts no new shared-file uploads. Existing files remain readable; shared text documents and skills have separate limits.
Where a workspace is billed, its storage and Channels charges come out of the credits of its billing owner, follow that owner's billing anchor, and are recorded and retried when credits are available. A member's own Free tier does not change the billing owner's allowance.
Check your balance
On code.foxl.ai, open Settings > Account: it shows the balance, the date your credits refill, the token breakdown, compute, and the ledger, each entry naming the task it was for. The desktop's Account page shows no balance; check it on code.foxl.ai.
What your own keys and sign-ins used through Foxl on your computer, by model, chat and day, is in the desktop app's Settings > Usage, under Usage Explorer (see AI Providers). Its costs are list-price estimates, not credits.
A positive balance allows a task to be considered for admission; each model request still needs enough available credits after existing reservations. Running out of credits stops new managed requests. A budget is not an advance debit: the ledger records usage as it is settled.
Troubleshooting
Managed inference is unavailable: Code cannot currently admit cloud model work. Retry when service is available; changing a Foxl Agent provider does not enable SaaS inference. This is an availability/policy response, not a reason to sign out.
Insufficient credits: inspect your available balance and pending work. Existing requests can hold credits, and the hold for a large input is sized above what it will cost (see Budgets and tiers), so a long conversation can be refused while some balance remains. Shorten the request or wait for the account's next grant.
Too many requests in flight: the account's managed requests (32 on Free) are already running. Wait for one to finish; a request that was cut off releases its slot within about 90 seconds.
Rate limit exceeded: a Free account has made 200 managed model requests in the last 24 hours. Requests are admitted again as the oldest leave that window.
No Foxl Agent provider connected: configure one under Settings > Model & provider. This message belongs to the local Foxl Agent path. Cloud Code uses its managed catalog instead.