Choose a model and read its usage
7 minute read · Chats & Local Agents
Choose a model for the current conversation after connecting its provider. Defaults help you start the next chat; changing a default does not rewrite the model choices in existing conversations.
- Choose what this task needs
- Keep the service tier deliberate
- Choose the provider account as well as the model
- Read usage without guessing
- Separate three different limits
- Change the model deliberately
- When usage looks stale
- Model picker and usage controls
- Continue with the accounts on this computer
- Choose a coding agent or a direct model
Choose what this task needs
Use the provider icon and model picker in the composer to check which coding agent and model the conversation will use. Select a discovered or supported custom choice. Reasoning effort, service tier and thinking options appear where the model and provider support them. A control available for one provider is not necessarily available for every provider.
Keep the service tier deliberate
Fast mode starts off in new chats. Enable it deliberately in a conversation only when the provider supports it and you want that service tier. Provider usage and billing rules still apply.
Linked tasks always use the standard service tier. Routine assignments use an available lighter model from the coordinator's provider and account; complex assignments can retain its stronger model when the coordinator supplies a reason. The agent cannot enable Fast mode for a linked assignment. See linked tasks.
Choose the provider account as well as the model
Provider accounts support separate named profiles for Codex, Claude Code, Google Antigravity, Grok Build, Kimi Code, Cursor, OpenCode, Muse Code and OpenRouter. Each provider keeps its own sign-in method and model capabilities. Give each profile a recognizable name and check its connection status. Your Vortex account, the provider, the selected account profile and the model are different choices; a provider icon alone does not identify which account will run the work.
Use the account controls to select the intended profile, then confirm the provider and model in the conversation before starting a run. Finish any required provider sign-in through the connection flow. Do not paste login tokens into chat or assume that adding another profile includes another subscription or more usage.
If the profile is signed out or unavailable, reconnect that profile and check its status before retrying. Available models and limits still depend on the provider account. Switching the visible profile is not proof that an already-running provider session has changed accounts.
Read usage without guessing
Conversation context and account allowance are different measurements. Context may be estimated. The ring in the composer shows the conversation's context percentage; open it for Context and account usage. The limits panel at the bottom of the sidebar, titled with the agent's name, shows the account limits for the current agent and model:
- Codex, Claude Code and Muse Code: 5 hours and Week, with the time until reset.
- Google Antigravity: up to four per-model quota rows, with the selected model first.
- OpenRouter: Balance, Key limit, Budget and Key usage when the key reports them.
- Limit reached with the reset time when a window is exhausted. Rows above 80% are highlighted.
Grok Build, Kimi Code, Cursor and OpenCode show Limits are not reported by this agent. when their protocol supplies no quota window, as does Muse Code until it reports values. This does not mean unlimited usage. Expired quota samples are not current balances.
Refresh the usage popover to inspect available values and errors. A provider eligibility or quota error belongs to that provider account; Vortex cannot change it. Vortex cloud storage is separate from your provider’s tokens.
Next: connect Codex, connect Claude Code, Vortex capacity.
Separate three different limits
| Display | What it answers | What it does not answer |
|---|---|---|
| Conversation context | How much of this model's context the current conversation uses; estimates are identified. | Your remaining paid account allowance. |
| Provider quota windows | The usage/reset information supplied by the connected provider. | How many Vortex projects you may retain. |
| Vortex cloud storage | Technical capacity for encrypted history, mail and signals. | A product count cap, included AI tokens or provider eligibility. |
Codex, Claude Code and Muse Code can report five-hour and weekly account windows. Google Antigravity exposes per-model account quotas and reset times, not conversation token counts in this integration. Grok, Kimi, Cursor and OpenCode show that limits are not reported where their protocol supplies no applicable windows. Never interpret absent information as an unlimited account.
Change the model deliberately
Open the model control in the current composer and select an available model. Review supported reasoning, thinking or service-tier settings where offered. A provider default and a conversation's saved selection are separate: changing a default should not silently rewrite earlier threads.
Before continuing an important run after changing provider settings, confirm the chosen agent and access mode. If a custom model is rejected, use a discovered supported choice and retain the error for diagnosis. Do not assume that a name accepted by one provider is valid for another.
When usage looks stale
Refresh the usage popover and read its update/reset information and errors. A sample collected before a quota reset should not be presented as a current restriction. If the provider returns a rate-limit error without a useful usage report, keep that error as the authority for the attempted run. Planning smaller tasks can reduce unnecessary context, but archiving a Vortex conversation does not reset an external provider's billing window.
Model picker and usage controls
| Control / indicator | Meaning |
|---|---|
| Provider icon / Agent tabs | Chooses the provider whose models/accounts you are viewing. Provider availability depends on installation and sign-in. |
| Account profile | Selects the actual provider account for the conversation. Work and personal accounts can have different models and quotas. |
| Search models | Filters the models in the current picker. |
| Favorites | Shows favorited models; favoriting does not grant access to a model. |
| Model availability | Displays the provider's available/known model information. |
| Refresh models | Re-reads model availability for the selected integration. |
| Custom model ID | Sends an explicitly entered model identifier through a supported provider path. Vortex cannot guarantee the provider accepts an arbitrary name. |
| Connection settings / Install agents | Opens the setup relevant to unavailable agents/accounts. |
| Thinking / Reasoning | Chooses supported reasoning effort. Higher effort can take longer and consume more usage. |
| Fast mode | Opts into a supported faster service tier for an ordinary chat. It starts off in new chats and is prohibited for linked tasks. |
| Context indicator | Shows available conversation-context information. It is different from a provider's periodic usage quota. |
| Sidebar limits panel | Shows reported limits/windows and reset information for the current agent and model. Missing or stale data is not unlimited allowance. |
| Model menu (Settings → AI providers) | Show … in the model menu chooses which providers appear for new chats across your account. Existing chats keep their provider. |
For linked tasks, keep the coordinating provider and account: a lighter available model is used for straightforward bounded work; complex tasks can justify the stronger model. Starting multiple chats still consumes usage even when every task uses the standard tier. See linked task controls.
Continue with the accounts on this computer
A synchronized conversation can refer to an account profile that does not exist here. Provider account choices are never synchronized. Vortex first uses an available local account from the same provider, regardless of its name. If that provider is unavailable, it uses another connected account on this computer. The fallback selection stays local and does not overwrite the other machine's choice.
Check the resolved provider and model before sending. If no usable account is connected, complete sign-in first. Explicitly selecting a provider that still needs connection opens that setup path instead of silently cancelling your choice. Already-running sessions keep the account they started with; another computer's provider session is not reused as a local login. See working between computers.
Choose a coding agent or a direct model
Codex, Claude Code, Google Antigravity, Grok Build, Kimi Code, Cursor, OpenCode and Muse Code are coding-agent integrations. OpenRouter connects directly, with its own account, API key, catalog and usage information. It does not need OpenCode installed. Its direct text chat uses explicit conversation inputs, without local agent tools.
Use model categories below search to find text, image, video, speech, audio, transcription and other supported models. Search and category filters combine. All models, or selecting the category again, clears that filter. Model metadata determines the settings and reference formats shown; provider availability and charges still apply.
Specialized OpenRouter models use Versions, with saved prompts, inputs and results. Selecting a version does not submit a request. Use inputs prepares a draft and Generate version submits it. A generation account, model and its saved input roles stay together when restoring that draft.