A: Hey CamInTalent,

That's a fantastic question about practical credit usage! It's tricky to give exact numbers for the demos because agent costs depend dynamically on the steps taken and tools used for each specific request.

Let's break down how credits generally work and how that applies to different agent types. Each step an agent takes has a fixed cost, which is 3 credits on Tiers 3 and 4, covering platform use. There's also a variable cost, but this mainly applies when using AgenticFlow's built-in resources, like our integrated LLMs (Gemini-flash, 4o-mini) or built-in image/video tools. If you use Bring Your Own Key (BYOK) for external services like OpenAI or Anthropic Claude, you don't pay the variable AgenticFlow credit cost for that specific external call; you only pay the fixed 3 credits for the step itself, and the actual AI cost is billed directly by the provider.

Now, applying this to the demo agents: Customer Success or Information agents like Ari or Rachel typically involve steps like checking knowledge bases and using LLMs for understanding and replying. If using our efficient built-in LLMs, these interactions are generally quite cost-effective in terms of credits.

Designer agents like Monica or Iris heavily rely on image generation. If using our built-in image tools, this will incur both the fixed step cost and a potentially higher variable credit cost. If you configure them to use an external service like Replicate via BYOK, the AgenticFlow cost is just the fixed credits per step, but you'll pay the external provider directly for each image generated, maybe around $0.01 or more depending on the service.

Deep Research agents almost always need a powerful external model via BYOK, like GPT-4. These involve multiple steps like planning, web searches, scraping, analysis, and synthesis. The AgenticFlow credit cost comes from the fixed credits for each of these steps. The main expense, however, is the external cost billed directly to your OpenAI account, which could easily be $0.50 or much more per research task depending on its complexity and depth.

So, the Tier 3/4 credit pools offer a lot of value, especially for workflows using efficient built-in tools or when shifting heavy LLM costs to external providers via BYOK. The best way to estimate for your specific needs is to try similar tasks and monitor the credit usage. A detailed usage dashboard is also coming soon to make this even clearer. Hope this helps!