☰

Guardrails for coding agents in safety-critical software.

Coding agents are already in your codebase. Code Sa maps your specifications, tests, source and downstream consumers into one knowledge graph, then proves, on every pull request, that the code still satisfies the specifications across the whole dependency chain, not just the file that changed.

BOOK A DEMO

READ THE ARGUMENT

RUNS ON

EVERY PULL REQUEST

LANGUAGES

C / C++

STANDARDS

ISO 26262-6 · MISRA C · MISRA C++

DEPLOYMENT

YOUR CLOUD TENANCY

WHAT CODE SA DOES

Four questions Code Sa answers on every pull request.

Evidence that regenerates at the speed the code changes, because that is the only cadence that now works.

Evidence that regenerates at the speed the code changes, because that
is the only cadence that now works. Read the argument →

Read the argument →

SPECIFICATION → TEST → CODE TRACEABILITY

Does the code still match the specification?

Verified

Partial

No coverage

—

Live status per requirement

across every approved statement, with the source text it came from.

—

Drill straight through

from a requirement to the exact tests that verify it and the lines that implement it.

—

Drift caught before merge

you see what a change breaks while it is still a pull request.

SPEC-001 — verified by 2 tests, implemented in 2 files, traced to libnimbus-v3-requirements.pdf p.3

SPEC-001 — verified by 2 tests,
implemented in 2 files, traced
to libnimbus-v3-requirements.pdf p.3

TEST SUITE ANALYSIS

Do your tests prove anything?

—

Every test decomposed

— the functions it calls, its assertions, and its size, at a glance.

the functions it calls, its assertions, and its size, at a glance.

—

Redundancy surfaced

— overlapping tests flagged as covered by another, so you can trim without losing coverage.

overlapping tests flagged as covered by another, so you can trim without losing coverage.

—

Gaps closed automatically

— targeted tests generated from real usage rather than from the code that was just written.

targeted tests generated from real usage rather than from the code that was just written.

Tests decomposed by APIs called, assertions and overlapping coverage

Tests decomposed by APIs called,
assertions and overlapping coverage

COMPLIANCE CHECKING

Does it meet the standard?

—

Every rule live

— pass or violated, with a running count of findings mapped to the ISO 26262 or MISRA clause.

pass or violated, with a running count of findings mapped to the ISO 26262 or MISRA clause.

—

Fixes proposed

— generated patches to bring non-compliant code back into line, for a human to accept or reject.

generated patches to bring non-compliant code back into line, for a human to accept or reject.

—

Re-verifiable on demand

— re-check the whole codebase against the guideline in one action.

re-check the whole codebase against the guideline in one action.

14 rules from ISO 26262-6 Tables 1, 3 and 6 — the list is itself the artefact clause 5.4.3 asks for

DOCUMENTATION GENERATION

Does the documentation describe what shipped?

—

Every API with its doc status

— generated, documented or undocumented, across the whole surface.

generated, documented or undocumented, across the whole surface.

—

Coverage scored per API

alongside how widely it is used, so the undocumented functions that matter get fixed first.

—

Regenerated on demand

from the current code, so the description and the behaviour do not diverge.

12 APIs with documentation state and per-API coverage, ranked by how widely each is used

12 APIs with documentation state and
per-API coverage, ranked by how widely
each is used

FREQUENTLY ASKED QUESTIONS

The questions that come up first.

Where does our source code go?

Code Sa deploys into your own cloud tenancy, so your source stays inside your perimeter and under your own access controls. Where a team wants to see output quickly and the codebase allows it, we can also run a managed instance on our infrastructure. The choice is yours and it does not change what the tool does.

Is our code used to train models?

No. Your source, your requirements and your test data are never used to train models — ours or anyone else's. When Code Sa runs in your own tenancy the question does not arise at all, because nothing leaves your environment in the first place.

GET STARTED

Adopt coding agents without giving up the evidence chain.

Tell us about your programme and the codebase you would want to run this against. We will come back with what the first pass would look like.

BOOK A DEMO

READ THE ARGUMENT

AUTOMOTIVE OEMS & TIER-1S

AUTONOMOUS DRIVING

SAFETY-CRITICAL EMBEDDED & ROBOTICS