The ExaktAI App

You ask in natural language, or using palettes of mathematical symbols, or describing the mathematics using Mathematica or Maple syntax. ExaktAI produces a succinct answer that includes the derivation steps, and automatically generates and opens a validation Mathematica or Maple document.

ExaktAI App showing the minimum distance to a plane solved using Lagrange multipliers, validated by Claude with the Mathematica notebook open below — ExaktAI running Mathematica as backend: problem validated step by step, with details collapsed. The Mathematica notebook was automatically generated and opened by ExaktAI.

Expanding any of the collapsed 'Details' shows, directly in the App, the computational instruction (CI) that represents the step in the Computer Algebra System (CAS), in this case Mathematica, the AI inferred result (AIR), and the result from the CAS. The CI and CAS are reproduced in the Mathematica notebook.

Expanded details for step 3 showing the computational instruction in Mathematica, the AI inferred result, and the Mathematica CAS result

ExaktAI App showing the osculating circle problem solved and validated in 4 steps by Codex with the Maple document open alongside — The Maple document was automatically generated and opened by ExaktAI; it mirrors the steps and can be re-executed or edited.

How it works

The problem is presented using natural language; ExaktAI guides the AI through a structured reasoning process that produces step-by-step computational instructions. Each instruction is executed in a CAS. ExaktAI applies a multi-stage strategy involving independent AIs, CAS executions, and mathematical equivalence checks in order to validate a step.

The validated answer appears in an automatically generated and opened Mathematica notebook or Maple document that you can audit, modify, experiment, and extend.

In summary: AI drives, CAS executes, ExaktAI orchestrates, and you stay in the loop.

AI consensus

Besides step-by-step validation, there is another mode: run several AIs on the same problem, then verify whether their answers are mathematically equivalent. ExaktAI uses computer algebra systems (CAS) to check equivalence and reports the agreement level, for example, Consensus 6/6 when six AIs independently converge on the same answer, and so on.

Consensus is not a substitute for full step-by-step validation. AI systems sharing the same training data can converge on the same wrong answer. What consensus adds is another way to assess an answer: disagreement is surfaced immediately, and convergence across independent AIs is a meaningful signal.

ExaktAI consensus mode: all six AIs (Claude, Codex, Gemini, DeepSeek, Mistral, Grok) independently produce the same answer for the osculating circle problem - Consensus (6/6) — Consensus (6/6) in 1'24": Claude, Codex, Gemini, DeepSeek, Mistral, and Grok all independently arrive at the same equation for the osculating circle. ExaktAI reports the shared answer and the agreement level.

When AIs Disagree

Consensus is not guaranteed. When the AIs split, ExaktAI shows which ones agree and explains the difference - automatically, by comparing the actual answers.

ExaktAI Split (3/6) on a vector projection problem. — Split (3/6) on a vector projection problem. ExaktAI groups the answers and explains the difference. (Math symbol palette hidden here for clarity.)

ExaktAI vs CAS AI assistants →

Request early access

ExaktAI is in development. We can reach out when it's ready to try, beta scheduled for late summer or fall 2026.