Juma Agent: AI model for marketing teams

When you open the model picker in Juma, you'll see something at the top that's a little different from the rest: Juma Agent, labeled "Most optimized for Juma." Below it are familiar names - Claude, Gemini, GPT. So what exactly is the Juma Agent, and why should you use it?

It's not just another model

The Juma agent isn't a single AI model. It's a smart layer built on top of the best AI models in the world - Anthropic Claude, OpenAI's ChatGPT, Google Gemini, Perplexity, Nano Banana, and others. Rather than locking you into one model, the agent automatically picks the right one for the right task, every time.

Think of it like a conductor in front of an orchestra. Each instrument (model) is world-class at what it does. The agent's job is to make sure the right one plays at the right moment - so the result is better than any single instrument could produce alone.

How the routing works

Different AI models are genuinely better at different things. Here's how the Juma Agent puts that to work:

• Writing and reasoning → Claude (Anthropic) for nuanced, high-quality content, and GPT-5 for extra reasoning and a different point of view
• Speed-sensitive tasks → Gemini Flash (Google) when you need fast responses
• Image generation → Nano Banana or GPT-image when visuals are involved
• Research and web-grounded answers → Perplexity for real-time, sourced information
• Code and technical work → Claude, optimized for accuracy and precision

You don't configure any of this. The agent handles it behind the scenes.

At its core, Juma Agent is an orchestrator

The most important thing to understand about it is that it's an orchestrator - a system that coordinates everything that needs to happen to complete your task, end to end.

When you send a message, a lot is going on underneath: figuring out what you actually need, choosing the right model, pulling in the right tools and integrations, managing context across multiple steps, and delivering a result that makes sense. The orchestrator is what keeps all of that working together - smoothly and invisibly.

A few of the things it handles that you'll never have to think about:

• Context trimming - keeping only the most relevant information in play so responses stay sharp and accurate
• Tool routing - deciding which integrations and capabilities to use based on your actual task
• Response compression - processing large outputs from connected tools so nothing gets lost or bloated
• Marketing-specific instructions - a layer of guidance baked in so outputs match how marketing people actually think and work

Why this matters for your team

Before the Juma Agent, using AI well meant knowing which model to pick - and switching manually depending on the task. That's a lot of overhead for a team that just needs to get work done.

Juma Agent removes that decision entirely. Whether you're running a multi-step research task, generating images for a campaign, writing a strategy doc, or connecting to HubSpot - the right model is already being used. You stay in flow.

It's especially built for:

• Long agentic workflows where multiple steps need to work together
• Marketing tasks like content creation, research, and campaign strategy
• Integration-heavy work with tools like Google Drive, HubSpot, Notion, and n8n
• Multi-step plans where accuracy and consistency across steps matters

Skills: purpose-built for specific tasks

On top of the model routing, Juma also has Skills - specialized instructions layered in for specific marketing tasks like creating social media carousels, building infographics, and generating presentations. Skills tell the Juma Agent not just what to do, but how a marketing professional would approach it.

This is part of what makes the outputs feel different from generic AI responses - the model is being guided by marketing expertise, not just raw capability.

When should you use a specific model instead?

Juma Agent is the recommended default and will handle most tasks better than picking a model yourself. However, if you have a specific reason - like testing a particular model, using a model your team is familiar with, or hitting a specific context window - you can always switch to Advanced models from the same picker.

The choice is always yours. The Juma Agent just makes the default a great one.