Skip to content

Make logic tangible

Write a rule.
See exactly what it does.

Tangibool is a workspace for drafting, checking, explaining, and testing Boolean logic. Start from plain English, a formula, or a diagram. Then check every case, read it back in plain words, and turn the case that breaks it into a test.

No account needed. It runs in your browser, and your rules stay on your device until you choose the cloud.

The Tangibool workspace showing the rule allow_entry := (member OR invited) AND NOT banned as a formula, a logic-gate diagram with live TRUE and FALSE values, a validation panel, a plain-English reading, and an evaluation trace explaining why the result is TRUE.
  • Every caseChecks are exhaustive or solver-backed, never a sample.
  • Works offlineUse it signed out; the engine runs in your browser.
  • 4 languagesWrite and read rules in English, French, German, or Spanish.
  • 13 file typesFrom Markdown and LaTeX to Python, SQL, and SMT-LIB.
  • 14 lessonsFrom NOT and AND to scales, counting, and strictness, open to everyone.

Try it here

Two ways to write one rule. Only one is right.

“Allow entry to members and invited guests, unless they are banned.” Your version and a teammate’s look almost identical. Flip the inputs and see whether you can find where they disagree.

Reading the formula won’t tell you. A diagram alone may not show it, and a plausible explanation isn’t evidence. Tangibool checks all eight cases, shows you the ones that differ, and lets you keep each one as a regression test.

  • Exact, not sampled. Every assignment is checked, or the result says “Not determined”.
  • Replayed before it’s shown. Each counterexample is re-run in the evaluator.
  • Kept as evidence. One click saves it as a test that guards the rule from then on.

How exact checks work

Entry rule · Simulation3 inputs · 8 cases

Inputs: select to switch

Results

Your ruleallow_entry := (member OR invited) AND NOT banned
TRUE
Your teammate’s versionallow_entry := member OR (invited AND NOT banned)
TRUE
Both versions give TRUE here. Agreeing on one case proves nothing, so try another or look at all eight below.

Every case, checked: both versions side by side. They differ in 2 of 8 cases.

memberinvitedbannedYour ruleTeammate’sAgreement
FALSEFALSEFALSEFALSEFALSE
FALSEFALSETRUEFALSEFALSE
FALSETRUEFALSETRUETRUE
FALSETRUETRUEFALSEFALSE
TRUEFALSEFALSETRUETRUE
TRUEFALSETRUEFALSETRUEDiffers
TRUETRUEFALSETRUETRUE
TRUETRUETRUEFALSETRUEDiffers

Start your way

Start from words, a formula, or a diagram

Every view edits the same underlying model, so the formula, the diagram, and the plain-English reading always agree with each other.

Start from English

Describe a rule in English, French, German, or Spanish. AI proposes a formula, a glossary, and open questions, with concrete alternatives when the wording is ambiguous. Nothing is accepted until you confirm it.

Start from a formula

Type or paste an expression in programming, mathematical, or textbook notation. Errors are pinpointed with suggested repairs, and your last valid version is never silently lost.

Start from a diagram

Add inputs and gates, connect ports, and choose an output. Cycles and missing connections are explained rather than accepted, and every editing action also works from the keyboard.

Check an argument

State premises and a conclusion. Tangibool tells you whether it follows, or shows the one assignment where every premise holds and the conclusion fails.

Beyond TRUE and FALSE

Numbers, categories, ratings, counts, and scores, checked exactly

Define each one outside the formula. Tangibool compiles it to TRUE/FALSE inputs, derives the facts between them, and keeps every check two-valued and exact.

Numbers and thresholds

“Age is at least 18” becomes age_at_least_18, defined as age ≥ 18. Give a number in a scenario and every threshold on that quantity is decided exactly, with the ordering between thresholds derived for you.

Categories and scales

Plans, countries, business activities, and ordered scales such as credit ratings or risk classes. Test membership with plan IN Paid or a level with rating ≥ BBB-, and get boundary and class tests proposed for you.

Counts and points

“At least two of three criteria” is one gate, AT_LEAST(2; …), not a chain of XOR. Points tables add up the rows that hold, and score thresholds are inputs your formulas use by name.

Several outcomes

Approve, Refer, Decline: define an outcome set as conditions, a rule list, or a decision table, and Tangibool checks that every case gets exactly one outcome, with a replayed case for each gap or overlap.

How these inputs work

Honest results

Five questions. Five separate answers.

A rule can be well formed and still wrong. It can pass your examples and still mean something you didn’t intend. Tangibool keeps these answers apart, so a green light on one never stands in for another.

When a check runs out of time, it says “Not determined”, never “passed”. Results from an earlier revision are marked stale. A simplification is called “simplified”, not “minimal”.

What “honest by design” means

  • Syntax

    Well formed

    Is the formula valid, and what exactly does it group together?

  • Mathematical check

    Equivalent

    Does the property hold for every allowed case? Exact, with a replayed witness.

  • Scenario tests

    3 passing

    Does it pass the examples that you, not the formula, said it should?

  • Interpretation

    Unconfirmed

    Does the formula say what the rule means? Only a person can confirm this.

  • Semantic review

    Optional

    Did an AI reviewer flag an omission or ambiguity? Never a mathematical check.

How AI fits in

AI drafts. The engine decides. You confirm.

Three kinds of evidence, never blended into one score. That’s the rule Tangibool is built around.

01 · Propose

AI suggests a reading

Drafting turns your words into a candidate formula, a glossary, and open questions. An optional reviewer looks for omitted exceptions or reversed conditions. Both are proposals, clearly labelled.

02 · Decide

The engine checks it

A deterministic engine evaluates, checks, and explains. It’s the same engine in your browser, on the server, in the command-line runner, and behind the API, and it never defers to an AI’s confidence.

03 · Confirm

You confirm the meaning

You confirm what the rule means and write the expectations. Change a word, a glossary meaning, or an assumption, and the confirmation is withdrawn until you look again.

Prefer no AI at all? Leave it off. Every deterministic feature keeps working, and nothing is sent anywhere.

The workspace

From first draft to reviewed handoff

Check it, test it, version it, and hand it over, without leaving the workspace or losing track of which revision you checked.

Exact checks

Find the case that breaks your rule

Ask whether two formulas are equivalent, whether a rule can ever be true, whether a conclusion follows, or whether one rule is at least as strict as another. You get a definite answer, or the exact assignment that proves otherwise.

  • Small models are checked exhaustively and cross-checked by a SAT solver
  • Larger ones use a bounded solver query and say “Not determined” if the budget runs out
  • Save any counterexample as a regression test in one click
  • Ask why a case got its result: the necessary conditions, the ways to qualify, and the smallest change that would flip it
The Compare tab: checking allow_entry against the candidate member OR (invited AND NOT banned) returns Not equivalent, with the counterexample member TRUE, invited FALSE, banned TRUE (allow_entry FALSE but the candidate TRUE), and buttons to load it into the simulation or save it as a regression test.

Tests

Catch regressions the moment they happen

Save the scenarios that matter, along with where each expected result came from. Edit the rule and rerun: a newly failing test is flagged as a regression, and the failing path lights up on the diagram.

After the formula is changed to member OR (invited AND NOT banned), the Tests tab flags 1 newly failing test since the previous run. The confirmed test expected allow_entry FALSE but got TRUE, and the failing path through the AND and OR gates is highlighted on the diagram.

Version control

Version every rule like code. No Git required.

Commit the formula, its source wording, glossary, diagram, and tests together. Branch to try an exception, compare text, structure, layout, and behaviour, then merge with conflicts shown rather than hidden.

  • Exact behaviour diffs with a witness: “before FALSE, after TRUE”
  • Layout-only changes are labelled as layout only
  • Restoring creates a new commit, so history is never rewritten
History compare between branch main and branch invitation-exception: logic structure changed with a text diff of the formula, behaviour changed with the witness member FALSE, invited TRUE, banned TRUE giving before FALSE and after TRUE, and before-and-after diagrams with the changed gates outlined.

Export

Hand off in the format they need

Export one artifact or everything from one pinned revision: Markdown and LaTeX reports, Python and PostgreSQL code, SMT-LIB queries, truth tables, and diagrams as PNG, JPEG, or SVG.

  • Generated Python, SQL, and SMT-LIB are cross-checked against the engine before every release
  • Exports never call AI and never change your model
  • Export All bundles a manifest with checksums for every file
The Export dialog: a revision selector with Snapshot only or Include authorised history, and format choices for each artifact: model report, source rule, glossary, formal model in .md, .txt, .py, .json, .sql, .tex, and .smt2, literal explanation, truth table, test suite, test results, and diagram images.

Explore every feature

Who it’s for

For anyone who has to get a rule exactly right

Developers

Refactor a tangled conditional and prove it still does the same thing, or get the input that breaks it. Generate Python, SQL, or SMT-LIB, and run the same tests in CI. Call the same checks from your own code through the API, or from an AI assistant through the MCP server.

Read the developer docs

Investment compliance analysts

Encode security-level eligibility rules over agency ratings, classifications, and issuer activities as precise logic. Show that an internal rule is at least as strict as the prospectus or the regulation, find the case that slips through, check each rule layer against the one above it, and keep a versioned record of every change for review. Limits computed elsewhere, such as a share of assets, come in as numbers tested against thresholds.

Product and business analysts

Turn a written eligibility, access, or workflow rule into logic you can read, test, and confirm with stakeholders, while the original wording stays right beside it. Model it over plans, tiers, counts, and scores with several outcomes, and show why a case got its result.

Students and educators

Learn operators, De Morgan’s laws, truth tables, Karnaugh maps, and counting with lessons that open real, checkable models, at beginner, intermediate, or advanced level.

Browse the lessons

Privacy

Private by default

Your rules may describe who gets access, credit, or a discount. They deserve care.

Local first

Use it signed out. Models, history, notes, and exports stay in your browser until you choose to copy a model to the cloud.

AI only with your consent

Nothing is sent to an AI provider until you switch that provider on yourself, and then only the text the feature needs.

No content in our logs

No product analytics. Our server logs never contain your formulas, notes, prompts, or variable names.

Read the privacy policy

Pricing

Free to start. Every plan includes every feature.

Use Tangibool free in your browser. Paid plans, from $9.99 a month, add cloud models, storage, and AI drafts. Viewers and commenters are always free, and verified students and teachers get 50% off Pro.