The idea is simple: know what your AI workflow is likely to cost before implementation.

I started working on this because in client work and while building another personal project, I kept running into the same problem: I couldn’t reason clearly about the whole system or share those ideas with others. Once retrieval, retries, tool use, infrastructure, and different model choices enter the picture, it becomes much harder to understand both cost and performance.

Flowcost is meant for that planning stage. It helps model AI systems before they’re built, so the architecture, tradeoffs, and likely costs are easier to understand and easier to share.

It’s not a monitoring tool or a live spend dashboard. It’s for planning AI workflows while the design is still flexible.

If you try it, I’d especially love feedback from engineers building LLM products, RAG systems, or agent workflows.