# Models & agent settings

Choose how your agent reasons and manage a supported model account connection.

Source: https://docs.opser.com/agents/models/

## Open the agent's General settings
Select your agent, then open **General**. The model list shows the models available in your environment, grouped so you can weigh capability against cost. Availability can change with your workspace configuration and any connected model account.

**Choose a model · Fictional data**

Advanced is selected from the Basic, Advanced, and Expert profiles. Custom opens the available models and supported thinking levels.


## Choose a model for the task
Models come from several providers — the Claude, GPT, and Gemini families, alongside other supported options — and are organized into three cost tiers:

| Tier | Best for |
| --- | --- |
| **Premium** | The most demanding reasoning: complex analysis, app building, and multi-step work where quality matters most. |
| **Standard** | Everyday work that still needs solid reasoning, at a lower cost per task. |
| **Budget** | High-volume or routine work such as short summaries and quick lookups. |

Each model also carries a capability label — **Basic**, **Advanced**, or **Frontier** — shown next to its name. This is separate from the cost tier, so you can spot a capable model that is still inexpensive to run. Pick the model in the selector, or open **Custom** to choose a specific model and thinking effort by hand. Regular OpenAI choices use Opser credits.

## Set the thinking level
The thinking level controls how much reasoning a model does before it answers: **Off**, **Minimal**, **Low**, **Medium**, **High**, and — on models that support it — **Xhigh**. More thinking can improve hard analysis but takes longer and uses more credits; less thinking is faster and cheaper for straightforward tasks. Not every model supports every level, so if you switch models the level may adjust to the closest one available.

## Give the agent a role
General also includes the agent's **personality** and **knowledge**. Personality describes how it should behave and communicate. Knowledge supplies domain facts and standing instructions for its role. These are separate from the organization's shared context and from the current conversation.

Choose the intended agent before editing. Each agent has one configuration, used for everyone who can use it; changing one agent does not change another. On an agent other people use, only its owner or an admin changes these settings. See [Context & memory](https://docs.opser.com/agents/context/).

## Connect a ChatGPT account
Open **Settings → Profile** and use **Linked accounts** to run a **Connect ChatGPT** flow that lets your agents use models from your own ChatGPT subscription. Start the connection, follow the sign-in steps, and return to Opser to confirm the connected state. Agents you own can access your account’s connection; a connection made earlier on a specific agent remains listed and manageable there. Never paste account credentials into a conversation.

Once connected, the model selector adds a second **Advanced** and a second **Expert** choice, each with a colored **Subscription** tag. The Custom dialog keeps regular GPT models in **OpenAI · Opser credits** and adds a separate **OpenAI · ChatGPT subscription** section. Only choices carrying the Subscription tag use your ChatGPT plan; the untagged Advanced, Expert and OpenAI choices continue to use Opser credits.

## Understand the cost boundary
A connected subscription and Opser credits are different billing arrangements. Model work runs on Opser credits unless it uses a connected account, which bills against that provider's own plan and limits. Connecting an account does not make every tool, integration, or generated artifact free — check the [Usage](https://docs.opser.com/billing/usage/) view for what a given kind of work consumes.

## Change or disconnect
Save your model and thinking choices in General. Use the **Linked accounts** controls in Profile to disconnect when you no longer want your agents to use a connected account. If a model stops being available, choose another from the list and check the account connection before retrying.