What does it cost to add a retrieval step to your AI pipeline? Most developers find out after they've built it - and often after the bill arrives.
Flowcost is a web tool designed to answer that question before you write the code. It lets you model the likely cost of a multi-step AI workflow by mapping out its components: retrieval (searching through documents or databases to give the model relevant context), retries (running the same call again when it fails or returns low-quality output), tool use (letting the model trigger external functions or APIs), and the specific models used at each step. Map those inputs, and you get a cost estimate at your expected usage volume.
Available at flowcost.ai, the tool is in early access. Pricing details aren't listed publicly yet.
The core problem Flowcost targets is that AI workflow costs compound in non-obvious ways. A workflow that makes 1,000 API calls might actually generate 3,000 if each call has a 50% retry rate and uses a retrieval step that makes its own sub-calls. Modeling that before building - rather than discovering it on your monthly bill - can change architectural decisions early, when they're cheap to change. A higher-quality model at twice the cost per token doesn't necessarily mean twice the total cost if it fails less often and requires fewer retries.
Where This Matters Most
The clearest use case is client work. Quoting an AI automation project requires knowing the estimated monthly operating cost. The alternative is quoting based on guesswork and apologizing later.
Internal build decisions are the other scenario: GPT-4o vs Claude vs Gemini often comes down to cost-per-thousand-calls at your volume. Flowcost lets you compare those scenarios before committing to a direction.
The inherent limitation: cost estimates are only as accurate as your usage assumptions. Volume projections, actual failure rates, and how often users trigger each workflow step all feed into the real number. Flowcost handles the math - the quality of your input assumptions is still on you.