Model access
Managed Code credits, Remote vendor keys, and desktop app provider access
Foxl Code and the Foxl desktop app have two distinct paths for model access.
| Where you run | Who supplies model access | Who bills model usage |
|---|---|---|
| code.foxl.ai orchestrator and managed coding agents | Foxl-managed Amazon Bedrock through Relay | Foxl account credits |
| Remote Claude Code, Codex, Hermes, and OpenCode | The same managed Code service | Foxl account credits |
| Remote Kiro and Cursor | Your vendor API key saved in Code settings | The vendor bills models; Foxl bills compute only |
| Desktop app local orchestrator and This computer agents | Your provider key, local model, or supported subscription (BYOK/BYOA) | Your provider or subscription; no Foxl model charge |
| A connected machine executing a managed Code task | Foxl Relay for models; your machine for compute | Foxl credits for models; your infrastructure bill for that machine |
In the Foxl desktop app, on app.foxl.ai and in the iOS app, Foxl Code takes the code.foxl.ai row by default (Remote), so its chat uses your Foxl credits. Set Where Foxl Code runs in Settings > Code > Coding agent to This computer for the Desktop app local row. See On your own computer.
A provider login in the desktop app does not become model access for code.foxl.ai. Managed agents use the managed catalog. Remote Kiro and Cursor require their own vendor API key saved explicitly in Code settings.
The website works with the Foxl app closed or uninstalled. It does not pair with the desktop app, discover local providers, or connect to a local terminal. Older local-machine defaults and saved desktop connection settings cannot send a website task to your computer; use the Foxl desktop app for local execution.
Managed orchestrator models
Choose the model beside the chat composer. The managed catalog offers these models:
| Family | Models |
|---|---|
| Claude | Opus 5.5, Opus 5, Fable 5.1, Fable 5, Opus 4.8, Opus 4.7, Opus 4.6, Sonnet 5.5, Sonnet 5, Sonnet 4.6, Haiku 5.5, Haiku 4.5 |
| OpenAI | GPT-6 Astra, GPT-5.6 Sol, GPT-5.6 Terra, GPT-5.6 Luna, GPT-5.5, GPT-5.4 |
| Z.AI | GLM 5.3, GLM 5 |
| Moonshot | Kimi K2.5 |
| xAI | Grok 4.7, Grok 4.6 |
| Gemma 4 31B |
Older versions can be hidden by your model-picker preferences while remaining available to saved selections. Models that support reasoning expose their supported thinking and effort controls.
Gemma 4 31B is chat-only in Code. It can answer and transform text, but cannot inspect repositories or launch coding agents. Choose a tool-capable model for orchestration.
The managed catalog is available to enabled accounts with credits, including Free accounts. Balance, request-concurrency, task, and repository-access limits still apply. A listed model is not a promise of unlimited usage.
Managed coding agents
The coding CLI determines which model family it can use:
| Agent | Managed models |
|---|---|
| Claude Code | Claude |
| Codex | GPT |
| Hermes | Claude |
| OpenCode | Claude |
Choose these under Settings > Code > Coding agent or in the task picker. No personal model credential is required. Backend availability also depends on which runtimes are provisioned for the environment.
Remote Kiro and Cursor
Kiro and Cursor are also available on Remote. Save the selected vendor's API key in Code settings, then choose the backend per task or as your default coding agent. A missing or unusable key is reported before Remote compute starts.
The vendor controls model access and model billing. Foxl charges compute only for these coding-agent runs. Your desktop app login is separate from the key saved for Remote.
What uses credits
Cloud chat, managed coding-agent model turns, and background orchestration or reviews use the Relay credit meter, and so does each Amazon Bedrock Web Search a Codex task runs. Remote compute is also metered. Kiro and Cursor model usage is billed by the vendor. Select a model and set task budgets before enabling autonomous or scheduled work.
Open Usage / Account in Code to see the balance, token usage, compute charges, and ledger. These remain visible even when purchasing plans or top-ups is unavailable. Credit accounting and checkout availability are separate.
If managed inference is temporarily unavailable, Code refuses new model work. It does not switch to your desktop app keys or subscriptions. Local use in the desktop app remains available through its configured providers.