Skip to main content

Documentation Index

Fetch the complete documentation index at: https://feasible-1447f9c5.mintlify.app/llms.txt

Use this file to discover all available pages before exploring further.

Feasible turns a problem described in plain English into a mathematical optimization problem and solves it. You describe what you want; Feasible figures out the variables, constraints, and objective, picks an appropriate solver, runs it, and explains the result.

​
A tiny example first

Before any theory, here is a very simple problem you can give to Feasible:
Tiny example problem
You have a $100 budget and you're picking items to take on a trip.
Each item has a price and a value to you.
Maximize total value, stay under budget, and you can take at most one of each item.

| Item     | Price | Value |
|----------|------:|------:|
| Tent     |   $40 |    50 |
| Sleeping bag | $30 | 40 |
| Stove    |   $28 |    25 |
| Lantern  |   $20 |    10 |
| Map      |   $10 |    10 |
Problems of this size can be solved by hand, and you will easily find that the solution is to take tent, sleeping bag, and stove for a cost of $98 and a value of 115. LLMs by themselves can also solve these tiny problems through reasoning. Here’s an excerpt from an actual answer from Claude:
[skipped intro text …] Greedy start — take items in efficiency order until the budget runs out:
  • Sleeping bag ($30, val 40) ✓ → $70 left
  • Tent ($ 40, val 50) ✓ → $30 left
  • Map ($10, val 10) ✓ → $20 left
  • Stove costs $28 — too much ✗
  • Lantern ($20, val 10) ✓ → $0 left
Greedy total: $100 spent, value = 110 Can we do better? The lantern is our weakest item (val 10, efficiency 0.50). What if we swap it out?
  • Drop lantern, keep stove? Tent + Sleeping bag + Map + Stove = $40+$30+$10+$28 = $108 — over budget ✗
  • Drop lantern entirely? Tent + Sleeping bag + Map = $80, value = 100. We have $20 left, but nothing fits (stove is $28) — value = 100 ✗ worse
What about skipping the map too?
  • Tent + Sleeping bag + Stove = $98, value = 115 ✓ Under budget, $2 to spare — nothing else fits
That’s better! Let’s check remaining combinations to be sure: …
The full answer can be seen here.
Problems of this type are called knapsack problems.

​
What about larger problems?

But what if the problem becomes bigger, with many more items to choose from? And several people are going on the trip together, each with their own budget? You will see that the number of combinations explodes quickly, and manually trying out becomes too complex to handle. In this case, you would want a mathematical optimizer to handle solving the problem. An very powerful optimizers for different kinds of problems do exist, both as open source and as commercial options. Unfortunately, they are not straightforward to use. You have to
  1. Model the problem as a mathematical optimization problem
  2. Install the correct type of solver
  3. Learn the solver API to transfer the model there and solve it
  4. Interpret the solver output correctly
The domain that deals with solving these kinds of problems is called Operations Research (OR).
Feasible is designed to help you with these steps. Think vibe coding, but for optimization problems.

​
What Feasible is

Feasible is an agentic interface for mathematical optimization. You describe a problem in natural language; Feasible:
  1. Understands the problem — what you are deciding, what you want to optimize, what the constraints are.
  2. Translates it into a mathematical program.
  3. Solves it using an appropriate open-source solver.
  4. Explains the answer back to you in plain language, alongside the numerical solution.
  5. Self-evaluates — Feasible re-checks its own answer and tells you when it isn’t confident.
You stay in control. You can ask clarifying questions, inspect the formulation Feasible built, edit the constraints, change solvers, or share the workflow with a colleague.

​
Main Features

  • Bring your own LLM key. Feasible is LLM-agnostic — it supports Anthropic, OpenAI, Google, and DeepSeek today, and is built so additional providers can be added without changing the workflow. You can use Feasible’s managed key (where available) or BYOK: store your own API key, encrypted, and Feasible uses it for your requests. No vendor lock-in. No silent provider switches.
  • Pick a provider per role, per request. Feasible runs two LLM roles in parallel — a worker that drafts the formulation, and a supervisor that reviews and challenges it. You choose the provider for each role independently, on every run. Strong reasoning model for the worker, cheaper one for the supervisor; or two different providers as a sanity check; or all-in on one. Your call.
  • Open-source solvers. The math is done by deterministic, open-source solvers. No black box, no proprietary cloud solver in the loop.
  • Full transparency. You see the variables, constraints, and objective the agent built, in human-readable form. You see the solver’s status and result. You see the agent’s self-evaluation. Nothing hidden.

​
What kinds of problems Feasible solves

Anything that fits the standard form of mathematical optimization — making decisions to maximize or minimize an objective subject to constraints. See the Worked Examples section on the left. Some typical applications:
  • Logic and combinatorial puzzles — N-Queens, Sudoku, knapsack, tiling.
  • Resource allocation — budgeting, portfolio selection, staffing.
  • Production and inventory planning — what to make, when, in what quantities.
  • Scheduling with custom rules and constraints.

​
When not to use Feasible

  • Free-form questions without a clear objective and constraints (“what’s a good marketing strategy?”) — Feasible needs something it can put in math.
  • Problems with thousands of variables — Feasible is best for small-to-medium models that the agent can reason about end-to-end. Large industrial models still benefit from a hand-written formulation.
  • Statistical or ML tasks — fitting a regression, training a classifier. Use a stats or ML library.

​
Where to go next

How Feasible works

The workflow under the hood: worker, supervisor, solver, self-evaluation.

N-Queens walkthrough

A logic puzzle that exercises the full pipeline on a verifiable problem.

Bakery production planning

A multi-period business problem with a quantitative objective.

Frontend tour

A panel-by-panel walkthrough of the Feasible app.