Your codebase
Any size. Any language.
Understand brings code exploration, graphs, metrics, standards checking, and AI-powered summaries together in one tool. Navigate unfamiliar code, see how components relate, measure code quality, and automate analysis across your workflow.
Runs entirely on your machine · No build required · 20,000+ developers
Understand is trusted by over 20,000 developers
Understand ships with a fully offline AI assistant. The difference isn’t the model — it’s what the model can see. Every answer is built on Understand’s parsed model of your codebase: the real call graph, every cross-reference, the dependency structure, the metrics, the standards violations.
Reads whatever source fits in its context window and infers the rest. Calls through macros, typedefs and function pointers go missing, and so does anything in a language it didn’t parse.
The real call tree, walkable in either direction, across language boundaries — every caller with a file and a line you can click.
Matches on the name. No way to separate a read from a write, and no way to tell you what it never saw.
Every reference in both directions, with reads separated from writes and each one located to a file and line. It works on code that doesn’t compile.
An opinion about your code, phrased confidently. A chat log is not evidence, and no auditor will take it as such.
3,000+ MISRA and AUTOSAR rules run per commit, deviations recorded, reported by a TÜV SÜD certified toolchain.
On whatever infrastructure runs the model, under whatever terms it was signed up under.
The model ships inside Understand and runs on your machine. Air-gapped installs included. Point it at your own local model if you’d rather.
It was never Understand or AI. Understand is where the AI gets its facts.
Understand parses your entire codebase and builds a connected model of its structure and relationships. Instead of searching through files manually, you can see how functions, types, modules, dependencies, and architecture fit together — even across languages.
Any size. Any language.
Parsed code model
Functions·Types·Variables
Calls·References·Dependencies
Architecture·Metrics·Standards
Once Understand has modeled the code, the questions that used to take hours become much easier to answer.
Each one is a thing engineers already do by hand — reading, tracing, guessing. Understand does them from the model.
Not what greps to a matching string — what genuinely reaches it, through macros, typedefs, function pointers, and across language boundaries. Invert the tree and you get everything downstream instead.
You get: a live call tree you can walk in either direction, exportable as an image or a report.
Hyper-Xref indexes every reference in both directions and separates reads from writes. Click the field, get the blast radius — including the module written in Ada that touches it through a binding.
You get: reads, writes, declarations, and definitions, each with a file and a line, on code that doesn't even compile.
State.theta14 refs · 4 filesDependency and architecture graphs drawn from the source, not from a diagram someone last updated three reorgs ago. Cycles and layer violations are flagged, and the same rules can fail your build.
You get: module coupling, cycles, and an architecture you can enforce in CI.
CodeCheck ships with 3,000+ rules — MISRA C 2004 and 2012, MISRA C++ 2008, AUTOSAR C++14 — and you can write your own in Python. The toolchain itself is TÜV SÜD certified, so the evidence counts.
You get: a per-commit violation report auditors accept, with recorded deviations.
Metrics at file, class, and function level — cyclomatic complexity, nesting, fan-in and fan-out, comment ratio — so a refactor gets scoped with numbers instead of the loudest opinion in the room.
You get: a ranked list of the code most likely to bite, and a diff against last release.
Every one is generated from the same parsed model of your code, not drawn by hand and never out of date. An AI chat window can describe your architecture in a paragraph; it cannot draw you the call tree, because it never resolved the references a drawing is made of.
3,000+ rules ship with every install. Write your own in Python. Run the whole thing in CI and the report lands in your inbox.
TÜV SÜD certified toolchain
ISO 26262 · IEC 61508 · EN 50128
Checked per commit against every MISRA edition listed below, with a traceable report per release.
MISRA compliance362 AUTOSAR C++14 rules, with deviations recorded where your project has signed off on them.
AUTOSAR complianceFive TQL-5 artifacts for DO-178C programs, with MISRA coverage and DER-ready traceability.
DO-330 qualification kitsAll of it installed by default — 3,000+ rules across the set. Checks you write yourself in Python appear in the same list, alongside these.
Dangling pointers, memory leaks, null dereferences, division by zero, unreachable paths — found by path analysis, not pattern matching.
How Bug Hunter worksThe same analysis engine, pointed at whichever of these is currently costing you the most.
Point Understand at the repository and it builds the map that was never written down: what calls what, which modules depend on which, and where the complexity that everybody avoids actually lives. No build system required — it parses source that hasn't compiled cleanly in a decade.
CodeCheck runs MISRA C 2004 and 2012, MISRA C++ 2008 and AUTOSAR C++14 against every commit, records deviations where your project has signed off on them, and exports a report traceable to file and line. The tool itself is TÜV SÜD certified, so it counts toward your qualification rather than adding to it.
Understand doesn't need the build to succeed, or at all. It parses what's there, resolves what it can, and tells you plainly what it couldn't — which is exactly the posture an independent reviewer needs. Then it documents what you verified in a form that survives cross-examination.
Define the layers you meant to build, and Understand checks the real dependency graph against them. Cycles, upward calls and boundary violations are flagged — and the same check runs headless in CI, so the build fails before the review does.
Before the deal closes, get the real picture: how the system is actually structured, how much of it is complexity nobody has priced in, and exactly which third-party code is bundled in and under what license.
Licenses start at $100 USD per month, billed annually with a twelve-month minimum. Tell us how many seats you need, which platforms you build on, and where your team sits — we’ll come back with pricing that fits.
Rather talk it through first?
Or call +1 (435) 627-2529 or email [email protected].
If yours isn't here, ask us directly — the answer comes from an engineer, not a form.
Talk to engineeringNo. Understand parses source directly rather than hooking a build. Partial, broken and uncompilable code still produces a usable model — which is the whole point for IV&V, litigation and inherited vendor code.
Yes, both. The analysis runs on your machine, and so does the AI: it ships with Understand rather than calling a hosted model, so there is no API key to obtain and no inference request to let out through the firewall. A fully air-gapped install is supported, with the AI working the same way it does anywhere else.
No. Analysis is entirely local, and so is the AI that answers questions about it. There is no cloud processing step for your code.
Yes. A CodeCheck rule is a Python script written against the same API the built-in rules use, so a check for your own internal coding standard runs, reports and fails the build exactly the way a MISRA rule does.
The CLI regenerates the database and runs checks headlessly. Most teams fail the build on new violations and publish the report as a build artifact.
TÜV SÜD certification for ISO 26262, IEC 61508 and EN 50128, with the qualification kit included in the standard license.
The complete product — every language, every rule, every graph. It is not a reduced edition.
A full-featured trial that installs in minutes and needs no build system. Or walk through your architecture with an engineer who has seen a few thousand of them.
No credit card required for the free trial. Available on Windows, macOS and Linux.