How Feasible works

The workflow that takes you from a plain-English question to a checked answer.

Feasible runs a workflow — a sequence of steps an agent walks through to take you from question to answer.

A few things to know:

• The agent does the modeling, not the math. It reads your problem, decides which variables and constraints are needed, and emits a structured representation. The actual optimization is performed by a deterministic solver.
• Worker and supervisor with self-evaluation. A worker drafts the formulation. A supervisor reviews and challenges it at different steps throughout the workflow. This catches a lot of small mistakes that come from naive single-pass generation. Each role can be backed by a different LLM provider — two different LLM working together typically reduces hallucination rates. You can set this on every run.
• Solver selection. Feasible has several optimization solver backends for different kinds of problems.
• Workflows are stateful and shareable. Every run produces a thread you can revisit, edit, or share. Further, you can download the model in different formats for further offline processing.

Pick the worker and supervisor providers

Each role can use a different LLM provider on every run.