MetaCortex·Dynamics
Products  /  MetaCortex Code
MetaCortex CodeGOVERNED AI CODINGv1.0 · controlled pilots

AI coding, with a governed path to merge.

For organizations that want the speed of autonomous coding without surrendering control of the repository. It converts AI-generated work into structured evidence: what changed, why, what tests ran, what failed, what repaired, what's blocked, and when the queue is actually closed.

It is not a chatbot in an editor. It is a governed code-production system.

Request a controlled pilot →See the governance model
EXPLICIT BOUNDARIES
proposal ≠ decision ≠ promotion ≠ execution
benchmark result ≠ authority  ·  status update ≠ terminal state
a generated patch is not progress · a passing benchmark is not authority · a green result is not closure
// OUTPUT IS NOT ADMISSIBLE CHANGE

AI coding agents can generate code, edits, tests, explanations, and status reports. They should not decide their own scope, certify their own evidence, approve their own completion, or declare their own release readiness. Most tools optimize for output. MetaCortex Code optimizes for admissible software change — change that carries evidence before it is allowed to count.

01  /  How it works
01
Bound the task
Starts with a declared scope — target files, change intent, success criteria. The AI generates candidates inside that scope. Nothing outside it is touched.
02
Generate with evidence
Every change is recorded: the task, target files, generated patch, tests, and evidence trail. The change is a candidate, not a decision.
03
Test and repair
Failures are not hidden. The system records failed checks, proposes bounded repairs, reruns evidence, and preserves the repair chain. If repair fails, it escalates — not suppresses.
04
Review with a package
Produces a human-reviewable package: what changed, why, what evidence supports it, what passed, what failed, what was repaired, what's open. The reviewer sees the full picture, not a summary.
05
Close on repo state
It does not stop because a status report was delivered. It stops when the queue is empty, the package exists, checks pass, escalation is clear, and the working tree is clean. Closure is a repo state, not an assertion.
02  /  Why it's different

Can the organization answer six questions about every AI-generated change?

If the tool cannot answer these with evidence, the change is not admissible — regardless of whether the tests pass.

Was the requested work bounded?
Did the change stay inside scope?
Was evidence produced?
Were failures recorded and repaired honestly?
Is the review package complete?
Is the repository actually in a closed state?
03  /  Proof points

v1.0 completed a bounded internal benchmark. Benchmark runs are evidence, not authority — full scores are not emitted unless full governed execution completes.

Bounded tasks completed10 / 10
First-pass CI rate9 / 10
Repair success1 / 1
False allows / false denies0 / 0
Replay-identical10 / 10
Hard-safetyfail-closed
What it is not
—a consumer coding toy
—an unrestricted autonomous merge bot
—a leaderboard-score generator
—a replacement for repository permissions or CI
—a claim that benchmarks are authority

It is a governed code-writing system for controlled deployment.

04  /  Who it's for
Engineering leaders
AI coding without uncontrolled repository mutation.
Regulated teams
Auditability, rollback, and evidence trails for every change.
AI infrastructure groups
Internal code agents that must stay inside governance boundaries.
Security-conscious orgs
No silent bypasses, unreviewed merges, or unverifiable output.
Safety-critical workflows
Every change traceable, repairable, and replayable.
Benchmark adapters
SWE-bench & Terminal-Bench — governed, bounded, evidence-only.
05  /  Specs
CategoryGoverned AI coding infra
Releasev1.0 achieved
InterfaceCLI / repo workflow
Stop disciplineRepo-state closure
Benchmark integrationSWE-bench · Terminal-Bench
AvailabilityControlled pilots
06  /  FAQ
Is it a coding assistant?

No. It's a governed code-production system. Generation is one step in a lifecycle of scoping, evidence, testing, repair, review, and repo-state closure. The governance is the product.

Does it replace my IDE or tools?

No. It operates at the repository level, not the editor. It integrates with your existing repo, CI, and review workflow, adding governance and evidence around AI-generated changes.

What does "repo-state closure" mean?

It stops when the repository is actually closed — queue empty, package exists, checks pass, escalation clear, working tree clean. The repository decides when work is closed, not the model.

What benchmarks does it support?

Protocol adapters for SWE-bench and Terminal-Bench. Runs are recorded, bounded, and governed; full score claims are not emitted unless full governed execution completes.

Can I use it with my CI/CD?

Yes. It does not replace CI/CD — it adds lifecycle governance around AI-generated changes. Your branch protection, test suites, and review processes remain in place.

Is it generally available?

v1.0 is achieved and available for controlled pilots. Contact us to discuss a pilot engagement.

Let the repository decide when it's done.

v1.0 is ready for controlled pilots.

Request a controlled pilot →All products