Current workflow: Discover all 49 analysis operations

Use the tool reference for input examples, computed outputs and method notes. Chat now supports topic-based tool discovery as well as exact tool names.

find tools for probability

Thirteen additional tools

text_profile, keywords, text_similarity, extract_summary, sentences, unique_lines, sort_numbers, moving_average, linear_fit, minmax_scale, shortest_path, word_definition and category_proof accept bounded typed inputs. Keyword counts and text similarity are lexical, not semantic understanding.

tool: moving_average {"values":[1,2,3,4],"window":2}

Use one result in another tool

A workflow references only earlier results. This example extracts source sentences, then counts content words in them.

workflow: {"steps":[{"id":"summary","tool":"extract_summary","input":"The orchard has apples. The orchard has trees."},{"id":"terms","tool":"keywords","input":"${summary.text}"}]}

Combine the existing calculators

All 30 bundled calculators can be called as builtin steps in workflows and reusable functions. The original direct tool commands remain available.

workflow: {"steps":[{"id":"area","tool":"builtin","input":{"name":"rectangle_area","args":{"width":4,"height":5}}},{"id":"total","tool":"calculate","input":"${area.value}+10"}]}

Learn combinations from examples

When a simple supported primitive does not fit, the learner can search up to 256 chains of one to three eligible one-argument functions from at most eight existing private skills. A composition is also considered before an underdetermined quadratic interpolation. Candidate construction uses training examples only; the reserved final example is a validation gate. Failure does not select another candidate using that held-out answer.

Versions and limits

Composed functions pin dependency versions. Existing regression examples must still pass on an update. Recursion, forward references and prototype access are rejected; shared execution budgets bound work. Example agreement does not prove intended meaning or correct extrapolation. There is no learned shell, network or arbitrary JavaScript execution.

Understand the output

Linear fit estimates association, not causation. Shortest path uses directed edges with non-negative costs. Category proofs use a selected dictionary sense. Saved user claims and lexical wording remain distinct from verified current facts.

Read the methods and limitations or try the examples in the workspace.