# Agent settings > Model, reasoning, access control, data handling, and the rest of an agent's configuration. Source: https://docs.runbear.io/agents/settings Last updated: 2026-10-07 The **Settings** tab holds the configuration that is not instructions, knowledge, or tools. Which sections appear depends on the agent's runtime; see [Runtimes](/agents/runtimes.md) for the full matrix. Opening the tab requires edit permission on the agent. Owners and Admins have it on every agent, and a member has it on an agent they were invited to under [access control](#access-control). Without it, the tab shows an access-denied page instead of the settings. ## Model The **Model Selection** section appears on Claude Agent SDK, Anthropic and OpenAI Responses agents; Classic agents have no model setting. Which rows it offers depends on the runtime: Claude Agent SDK agents offer Claude, GPT and Gemini models, Anthropic agents offer Claude models, and OpenAI Responses agents offer GPT models. | Model | Best for | Relative cost | | --------------------- | ---------------------------------------------------------------- | ------------------------------------------------------- | | Claude Fable 5.1 | Long-horizon agentic work, multistep research, hardest reasoning | Highest, and billed by Anthropic rather than in credits | | Claude Opus | Complex analysis, research, nuanced writing | Highest | | Claude Sonnet | General-purpose work, coding, problem-solving | Medium | | Claude Haiku | Quick answers, high volume, routine operations | Lowest | | GPT Astra | Complex analysis, multistep research, hardest reasoning | Highest | | GPT Sol | General-purpose work, coding, problem-solving | Medium | | GPT Terra | Balanced capability across most work | Medium | | GPT Luna | Fast, inexpensive, latency-sensitive work | Lowest | | GPT-5.6 Sol | The strongest reasoning of the GPT-5.6 tiers | Highest | | GPT-5.6 Terra | Balanced capability across most work | Medium | | GPT-5.6 Luna | Fast, inexpensive, latency-sensitive work | Lowest | | Gemini 3.1 Pro | Long-context reasoning and agentic workflows | Highest Gemini cost | | Gemini 3.8 Flash | Fast agentic workflows, coding, and high-volume work | Medium Gemini cost | | Gemini 3.5 Flash Lite | High-volume routine and latency-sensitive work | Lowest Gemini cost | The Claude Opus, Sonnet and Haiku rows are named by family, and on Claude Agent SDK agents the GPT Astra, Sol, Terra and Luna rows are named by tier. Each carries a **Stable** badge: a Stable row automatically follows the latest available version in that family or tier. Today the GPT tiers follow GPT-6 for Astra, Sol and Luna, and GPT-5.6 for Terra. To keep the agent on one exact release, choose it under **Specific version** instead. Specific versions may become unavailable when their provider retires them. The GPT-5.6 Sol, Terra and Luna rows are the OpenAI Responses agent's tiers; that runtime has no Stable rows, and lists older GPT versions under **Other versions**. A model the agent can't use right now stays in the list, shown disabled with a badge that says why: | Badge | Meaning | | -------------------------- | ------------------------------------------------------------------------------------------------------ | | **Requires Own API Key** | The model runs only on the agent's own provider key, not on Runbear-provided access | | **Requires Runbear's Key** | On a Claude Agent SDK agent using its own key, the model belongs to a different provider than that key | | **Unavailable in HIPAA** | The model can't be used while the agent is in HIPAA data handling | For Claude Agent SDK agents, Runbear-provided access uses Runbear credits. If the agent uses its own API key, Claude models require an Anthropic key, GPT models require an OpenAI key, and Gemini models require a Google AI Studio key; the selected provider bills usage on your own key. To change providers while using your own key, first switch the agent back to Runbear-provided access, select the new provider's model, then save its key. Claude Fable 5.1 runs only on an agent's own Anthropic key. It is not offered on Runbear-provided access, so until the agent is on its own key the row is shown disabled with the **Requires Own API Key** badge. A more capable model is not automatically a better agent. If an agent mostly answers from a knowledge base, the cheapest tier usually matches the most expensive one on answer quality while consuming a fraction of the credits. See [billing and credits](/administration/billing-and-credits.md). ## Reasoning The **Reasoning** section appears only on OpenAI Responses agents. **Effort** sets how much the model thinks before answering, and **Verbosity**, shown for models that support it, sets how much of that it writes out. Both start at **Default**, which uses OpenAI's own setting. Higher effort improves multi-step work and tool sequencing, and costs more credits per message. Leave it low for lookup-style agents. ## API key The **API Key** section appears on Anthropic, Claude Agent SDK and OpenAI Responses agents. By default the agent runs on Runbear's key and uses Runbear credits. To use your own key instead, turn on the switch named for the provider (**Use my own Anthropic API key**, **Use my own OpenAI API key** or **Use my own Google AI Studio API key**), paste the key, and click **Save**. Anthropic agents use an Anthropic key and OpenAI Responses agents an OpenAI key; Claude Agent SDK agents use the key of the selected model's provider: Anthropic for Claude models, OpenAI for GPT models and Google AI Studio for Gemini models. With your own key, the provider bills your account directly and the agent uses no Runbear credits. A saved key is never displayed again; submit a new one to replace it. Turn the switch off to return the agent to Runbear's key. Changing the key requires the `feature:write` permission on the agent. If an OpenAI Responses agent's uploaded files live in a vector store on your own OpenAI account, it can't switch back to Runbear's key from here; the section says so and points you to the [Classic Editor](/agents/testing.md#the-classic-editor), which re-provisions the store. ## Access control Access control decides who can view and edit this agent in the dashboard. Choose one of: - **All organization members can view and edit** — the default. Every member can open the agent, but changing it, including its **Settings** tab, still needs write permission on the agent; see [Members and roles](/administration/members-and-roles.md). - **Only invited users can view and edit** — then add the members who should have access. This setting does not affect who can talk to the agent. Anyone in a connected Slack channel, Microsoft Teams chat or other channel can still use it there. For Slack, a connection can be limited to specific users where that option is available; see [Restricting who can use a Slack connection](/channels/chatbot-settings.md#restricting-who-can-use-a-slack-connection). ## Data handling The **Data handling** section appears only for organizations entitled to HIPAA mode. It sets whether the agent uses standard data handling or HIPAA data handling. In HIPAA mode the agent runs on a restricted processing path; agents left on standard data handling are unchanged. Only Anthropic agents can be switched to HIPAA. The mode is chosen per agent, and it appears in the create dialog as well, so a new agent starts in the right mode instead of being switched afterward. A HIPAA agent can be returned to standard data handling: select **Standard**, click **Save data handling**, and confirm **Return to Standard** in the "Return this agent to standard data handling?" dialog. Don't do this while the agent handles protected health information. Claude Fable models are not available in HIPAA mode. An agent in HIPAA mode cannot be moved onto one, and an agent already on one cannot be switched into HIPAA mode. Before switching modes, read the [HIPAA setup guide](/security/hipaa.md) for Enterprise eligibility, BAA requests, supported runtimes, knowledge migration, and data-handling limits. ## Long-term memory The **Enable long-term memory** toggle ("Remember user preferences and frequently referenced resources across conversations") is on by default. What it controls depends on the agent's runtime; see [Memory](/agents/memory.md). ## Response feedback **Enable response feedback** adds feedback buttons under this agent's Slack responses ("Show 👍 / 👎 feedback buttons under this agent's Slack responses"), so people can rate an answer with one click. It applies to Slack only and doesn't use emoji reactions. The **User Feedback** chart in [analytics](/administration/analytics.md) is separate: it counts the emoji reactions people add to the agent's replies. ## Tool activity While an agent works, it reports each tool call as it happens — a task card in Slack, and a `thread.tool_call.progress` event on the API and SDK streams. The **Tool Activity** section controls whether this agent does that. The setting is on by default, and turning it off is a presentation change only: the agent still runs exactly the same tools, and the run's own trace still records every one of them, so the activity remains available to anyone reviewing the run afterwards. What stops is the live reporting, on every surface at once — a reply arrives without the intermediate steps. Turn it off when the steps are noise to the people reading, or when a custom frontend built on the API should show only the answer. The section appears only on agents whose model family reports tool calls at all. Where it is absent, the agent has no tool activity to hide rather than a setting that was withheld. Surface still matters independently: Teams and Discord replies never show tool calls, whatever this setting says. ## Choice buttons Rather than ending a turn with a typed-out question, an agent can offer the answers as clickable options — "Approve" and "Reject", or a short list of the choices it was about to describe. The **Choice Buttons** section controls whether this agent may do that. The setting is on by default, and turning it off withdraws the buttons on **every** surface the agent answers on, not just one. Turn it off when the agent should carry out what your instructions authorize without stopping to ask. The section appears only on agents whose runtime reads it. Native Gemini, Perplexity, Upstage and legacy OpenAI Assistants agents never offer buttons, so the section is hidden there rather than shown as a switch that does nothing; a Claude Agent SDK agent keeps it whichever model it runs. Buttons are also never offered on scheduled or other automated runs, where nobody is waiting to click. Clicking an option runs the choice as the person's next turn, so it is billed like any other message. The conversation is never blocked on a click — answering in your own words instead is always valid. Where the buttons appear, and how long they stay clickable, depends on the channel. Slack renders them as message buttons and deliberately **leaves them in place after a selection**, so someone can try another option later; each of those clicks runs as its own turn. Rendering them in the [Web SDK](/api/web-sdk/response-components.md) is in limited availability — Runbear enables it per organization, and it is then configured per agent, so the switch above being on is necessary but not sufficient there — and there the buttons go inert once a later message answers them. Where it is enabled, the switch reaches further than its name: it also governs the **card**, a formatted block of details the agent displays without asking anything. Turning choice buttons off withholds both. How often the agent offers them there is a separate per-agent dial, **Component frequency**, which sits under the switch in this section with the levels **Off**, **Low**, **Medium** and **High**. It appears only once Runbear has enabled it for your organization, and only on agent types that can emit components — Anthropic, OpenAI Responses, Mastra and Claude Agent SDK agents. Changing it requires the `feature:write` permission; without it the select is shown but disabled. The switch above needs no such permission. The dial affects only the Web SDK, and it stays visible and editable while choice buttons are off, but it does nothing then: the switch is checked first. The same level can be read and set over the REST API — see the [Response Components API](/api/response-components.md). ## Related - [Runtimes](/agents/runtimes.md) - [Instructions](/agents/instructions.md) - [Billing and credits](/administration/billing-and-credits.md) - [Members and roles](/administration/members-and-roles.md)