Introducing Rovo Agent Reasoning: choose how your agents think

[#1]

Choose how your Rovo Agents think with Agent Reasoning

Hello Community 👋

Not every agent task deserves the same amount of thinking.

A quick policy question should be answered quickly. But investigating a complicated customer issue, weighing competing options or bringing together context from several sources calls for more. Agent Reasoning gives you a practical way to match your agent’s effort to the work in front of it.

That means you can keep everyday agents fast and focused, while giving more demanding work the extra reasoning capacity it needs. It’s control over both the depth of the work and the model powering it.

Meet the reasoning tiers

Quick Answers is the default for Rovo Agents. It is fast, balanced, and built for the work that fills most days: answering questions, summarising content, finding information and routing requests.

Think Deeper gives your agent extra capacity to work through more involved problems. Use it when the job involves complex multi-step execution, weighing/prioritising options, connecting context across many sources and producing rich quality content.

Within Think Deeper, you can currently choose from available models including Gemini Flash 3, Gemini Flash 3.5, GPT 5.4, GPT 5.6 Luna, Claude Opus 4.6 and Claude Sonnet 4.6. The available options may evolve over time, so check the Agent builder for the latest selection.

What this unlocks

• Right-sized effort: keep straightforward tasks quick without underpowering complex work.

• More capable agents: give multi-step analysis and cross-source work the time and reasoning depth they need.

• More control when it matters: choose a model within Think Deeper to tune deeper reasoning for your use case.

Set it up in a few clicks

1. Open an existing agent - or create a new one - in the Studio.

2. Find the Reasoning field in the agent configuration.

3. Choose Quick Answers for fast work or Think Deeper for more involved tasks.

4. For Think Deeper, optionally choose the model that best fits your needs.

5. Save and test. You can change the tier at any time, and the update takes effect immediately.

 

Where should you start?

Start with Quick Answers. It is the fastest option and a strong fit for the majority of everyday tasks. If your agent needs to analyse a problem more deeply, pull together several sources, or produce a more thorough result, try Think Deeper and compare the outcome.

Think Deeper applies more reasoning and uses more credits, so it is best used intentionally-especially for high-volume or automated agents. Match the tier to the quality and depth the task requires.

We’d love your feedback

Agent Reasoning is built to give you meaningful control without making agent building more complicated. Try the tiers with your own agents, tell us what changes, and share the use cases where deeper thinking makes the biggest difference. Questions, ideas or unexpected results? Add them in the comments below 👇

Thanks for reading,
Sunny

[Martin Runge]

Great to see this land here! Reasoning tiers are already standard in most AI tools, so the mental model transfers straight over. What I was looking for here is the cost side, and the article does answer it: Think Deeper uses more credits, with a warning for high volume or automated agents. So the tier is not a token saver on its own. Quick Answers is the saving; Think Deeper is the spend.

Still open for me: is the credit cost per run higher regardless of output length, do the models under Think Deeper differ in cost from each other, and can admins see consumption per agent to catch a runaway automated one early?

My practical approach would be to leave everything on Quick Answers, then promote single agents to Think Deeper only after a side-by-side test on a real prompt shows a better result.

Cheers, Martin

[Sunny Lin]

Hey @Martin Runge

Glad to see the excitement for this and agreed, the mental pattern should be similar. Definitely makes sense to default to Quick Answers and then apply Think Deeper (or other models) on certain use cases - Quick Answers with Rovo is definitely still powerful itself.

On the cost questions:

• Credit costs on Think Deeper aren't always higher than Quick Answers depending on the prompt - if it's a simple prompt at execution, Think Deeper charges the same to Quick Answers
• Yes the costs of the models do differ and we provide an indicative multiplier there for the models
• Not currently but we're quickly aiming to provide a per-agent cost view soon to admins!

Thank you!

Sunny