---
title: "Agent settings and your own OpenAI key"
description: "Set the agent's reply language, tone and web search, manage what it remembers, run it on your own OpenAI key, and top up your AI budget."
url: https://help.rankprompt.com/en/articles/71-agent-settings-and-your-own-openai-key
collection: "Agent"
updated: 2026-11-15
language: en
---

# Agent settings and your own OpenAI key

Set the agent's reply language, tone and web search, manage what it remembers, run it on your own OpenAI key, and top up your AI budget.

**Agent settings** holds how the agent replies, what it remembers and your own OpenAI keys. Open it from the gear at the foot of the sidebar, under **Settings > Agent settings**, or from the settings menu inside any agent chat. The page is open on every plan.

The GitHub connection the agent uses to open pull requests is not here: set it up on the GitHub card in **Integrations > Publish content**.

* * *

## Replies

<table style="min-width: 50px;"><colgroup><col style="min-width: 25px;"><col style="min-width: 25px;"></colgroup><tbody><tr><th colspan="1" rowspan="1"><p>Setting</p></th><th colspan="1" rowspan="1"><p>What it does</p></th></tr><tr><td colspan="1" rowspan="1"><p><strong>Reply language</strong></p></td><td colspan="1" rowspan="1"><p>The language the agent answers in: pick any language, and search the list by name. <strong>Match my messages</strong> (the default) answers in the language you write in.</p></td></tr><tr><td colspan="1" rowspan="1"><p><strong>Tone</strong></p></td><td colspan="1" rowspan="1"><p>How much the agent explains: <strong>Balanced</strong> (a clear answer with a few short lines), <strong>Concise</strong> (the answer and little else) or <strong>Detailed</strong> (the reasoning and the steps)</p></td></tr><tr><td colspan="1" rowspan="1"><p><strong>Web search</strong></p></td><td colspan="1" rowspan="1"><p>Lets the agent look up current facts about your brand, your competitors and AI search</p></td></tr></tbody></table>

Things the agent saves for a brand stay in the brand's market language, whatever the reply language.

* * *

## Memory

**Memory** lists what the agent remembers about you and your brands, newest first. Click **Forget this** to remove one. See [Agent memory](https://help.rankprompt.com/en/articles/68-agent-memory.md).

* * *

## Agent API keys: run the agent on your own OpenAI key

By default the agent spends your AI budget. With your own OpenAI key, the agent runs on your OpenAI account instead: its usage is billed by OpenAI and your AI budget is not touched. That includes the help center search the agent runs before most answers, which goes to your OpenAI account with the answer. Agent API keys need Starter or higher on your own account.

### Adding a key

1.  Go to **Agent settings > Agent API keys**
    
2.  Click **Add**
    
3.  Enter a **Name** (a label to recognise it) and the **API key** (OpenAI keys start with "sk-")
    
4.  Click **Add key**: the app checks the key with OpenAI before saving it
    
5.  Click **Activate** on the new key
    

A new key starts inactive. Only one key is active at a time. You can store up to 5 keys.

### Choosing a model

Each key has a model picker with the models your key can run the agent with. One is marked **Recommended**.

### Switching back to the AI budget

Click **Deactivate** on the active key, or delete it. The agent goes back to your AI budget.

If your key fails during a conversation (expired, out of quota), that turn runs on your AI budget and the chat says so.

Your key is stored encrypted. After saving, the app only shows a masked hint of it.

* * *

## AI budget

Without your own key, the agent spends **AI budget**: a dollar balance separate from credits, included in every plan ($1 a month on Free, up to $25 on Agency Plus; see [What is the agent?](https://help.rankprompt.com/en/articles/70-what-is-the-agent.md)). AI edits in the article editor spend it too.

### Checking it

-   **/usage** in the chat shows the conversation's context and the AI budget used, which inside a teammate's brand is that brand owner's
    
-   **Subscription > Usage** shows the AI budget with every credit pool, and the agent's spend under **AI usage**: each answer with its model, tokens and cost, and the help center search it ran with it
    

An answer's cost includes that help center search. The agent starts a reply only while the budget left covers a long answer, a few cents, so it can say the budget ran out before the balance reads exactly zero.

### Topping it up

Convert credits into AI budget on **Plans > AI budget**, or from **Top up your AI budget** when the chat tells you it ran out:

<table style="min-width: 50px;"><colgroup><col style="min-width: 25px;"><col style="min-width: 25px;"></colgroup><tbody><tr><th colspan="1" rowspan="1"><p>Credits</p></th><th colspan="1" rowspan="1"><p>AI budget</p></th></tr><tr><td colspan="1" rowspan="1"><p>20</p></td><td colspan="1" rowspan="1"><p>$5</p></td></tr><tr><td colspan="1" rowspan="1"><p>50</p></td><td colspan="1" rowspan="1"><p>$15</p></td></tr><tr><td colspan="1" rowspan="1"><p>120</p></td><td colspan="1" rowspan="1"><p>$40</p></td></tr></tbody></table>

AI budget you add is spent within the current billing cycle and resets with your plan. If you do not have enough credits to convert, buy a credit top-up first. See [Plans and subscription](https://help.rankprompt.com/en/articles/65-plans-and-subscription.md).

* * *

## FAQ

### Does my own key use my AI budget?

No. While a key is active, the agent's usage goes to your OpenAI account. The one exception is a turn where your key fails: that turn runs on your AI budget.

### Does my key work for my teammates?

No. Keys and memories are per person. The AI budget follows the brand, the way credits do: whatever a teammate does inside your brand (agent chats, AI edits in the article editor, Community drafts) spends your AI budget as the brand owner, and the chat shows them your budget. A teammate with their own active key runs the agent on that key instead. Only the owner can top the budget up.

### Why did my conversation end?

It grew too long. A conversation has a limit on how much it can hold: close to it, the chat shows how much of its context is used and offers **New chat**, and past it the conversation takes no more messages. This happens on your own key too. Nothing is lost: the conversation stays in your history and whatever it created or changed is saved. Start a new chat to carry on.

### Which model does the agent use without my own key?

A model Rank Prompt picks for quality and speed. Its answers spend your AI budget; actions that cost credits, like running a report, ask first.

* * *

## Next steps

-   [What is the agent?](https://help.rankprompt.com/en/articles/70-what-is-the-agent.md)
    
-   [Using the agent](https://help.rankprompt.com/en/articles/73-using-the-agent.md)
    
-   [Agent memory](https://help.rankprompt.com/en/articles/68-agent-memory.md)
    

* * *

## Questions?

Email [support@rankprompt.com](mailto:support@rankprompt.com) and I'll help.
