Continuous Design Audit

Design Continuity
for Agentic Engineering

AI tools generate production code faster than architects can govern it.

rkito continuously enforces design intent on every pull request— keeping AI-generated software conformant before it merges.

Commitment to open source

Free for open source.

Qualifying OSS projects get the full platform, for life.

For everyone else

A free tier, always.

Start now on Starter — no credit card needed.


The challenge

AI Changed the Bottleneck

Ballooning token budgets. Slipping quality.
Agentic engineering that doesn't scale.

Agents generate at a pace no review process was built for. Left unmanaged at that scale, it threatens business continuity — and three problems become starkly visible.

Engineering leaders are thinking

01Why are my AI bills exploding?
02Why are my architects drowning in reviews?
03Why does every AI-generated PR require manual validation?
04Why does quality keep dropping the more I scale agentic engineering?
05Why is architectural drift increasing?
06How do we instill and ensure trust in AI-generated code?
01 · Ungoverned intent
Scattered intent.
Bigger token bill.

Today, design intent lives everywhere: READMEs, ADRs, CLAUDE.md/AGENTS.md files, wikis, and design documents. Governance, if it exists at all, relies on review workflows built for a handful of human pull requests each week—not continuous, agentic engineering at scale. Every additional source expands an agent's context, increasing token costs with every prompt. Yet even after paying that cost, the agent is still left to reconcile conflicting or incomplete intent. Without a single source of truth, design conformance degrades and architectural drift compounds with every generation.

02 · Confidence gates rollout
The Agentic Engineering
Bottleneck

Engineering leaders want to accelerate agentic software development—not at the expense of delivery quality. As AI-generated code becomes a larger share of every release, design inconsistencies, review overhead, and architectural drift become increasingly difficult to control. Senior engineers and architects are pulled into reviewing larger and more frequent pull requests, creating a bottleneck that limits scale. Without automated design conformance, organizations cannot confidently increase the autonomy of their engineering agents.

03 · The constraint moved
Can't shift left.
Can't shift right.

Most delivery pipelines were never built to incorporate agentic engineering, let alone the CI/CD shape it demands. Teams struggle to shift left—catching design intent early, wherever agents or AI-enabled ideation are involved—and struggle just as much to shift right, catching drift at the speed agents now generate. An engineering organization that can't execute both cannot transform for the agentic era.

How It Works

Design Continuity at Scale
Automated. Continuous. Frictionless.

rkito eliminates the operational friction of governing agentic engineering. From maintaining design intent and steering artifacts to declaring change intent, measuring conformance, auditing architecture, and detecting drift, rkito automates design governance directly within the GitHub pull request workflow—without changing how developers work.

01
Steering

Steering
Artifacts

Where design intent is authored — as Concepts, Policies, Best Practices, and Patterns. Every artifact is AI-assisted to write, so design thinking can diverge and converge quickly — but every change still passes a human-only review before it's approved. Each artifact ultimately resolves into Lenses: micro-contexts with a binary, yes/no evaluation outcome — the atomic unit the CDA actually evaluates.

ConceptsPoliciesBest PracticesPatterns→ Lenses
+
02
Architecture Ledger

Architecture
Ledger

The living, current, approved state of every system, component, and interface in your architecture. Ledger artifacts are authored the same AI-assisted way as Steering artifacts — but they describe what actually exists, not what should be true. This is the baseline every future Change Intention is proposed and audited against.

SystemsComponentsInterfacesBoundaries
+
03
Change Control · CHI

Change
Control

Ledger artifacts change through exactly one controlled path: a Change Intention (CHI). A developer — or an agent — declares what they want to do; rkito captures it as a CHI and proposes how the Ledger should change as a result. Every CHI goes through Design Review, a human-only gate much like a GitHub pull request. Once approved, it merges into the main branch of design.

Change Intention (CHI)Design ReviewMain Branch of Design
+
04
Continuous Design Audit · CI/CD Gate

Continuous
Design Audit

Every pull request automatically triggers a Continuous Design Audit.

rkito identifies the applicable design intent, evaluates conformance, explains every violation, and blocks unapproved architectural drift before merge.

Pass
conformant
Record
approved drift
Block
unplanned drift
Design continuity strategyafter code is written · before it merges

Shift
Right.

WritePushBuildLintTestsrkitoCDAenforced here

No architect can hold all design intent in memory — especially at this pace.

rkito is the institutional memory that never sleeps. Every PR. Every merge. Fully automated.

rkito empowers agentic engineering
Steering
Design
Intent

ADRs, Policies, Lenses — the rules the system must obey.

Copilot · Cursor · Claude
AI Agent
Generates

Syntactically correct. Intent-blind. No awareness of constraints.

The missing gate
rkito
CDA

Design intent enforced on every PR. Automatically. Before merge.

Outcome
Ships,
governed.

Conformant. Auditable. Drift caught before it compounds.

The Core Concept — Continuous Design Audit

Every CI/CD pipeline
has a gate missing.

Agentic environments that write and ship production code need a design intent gate. Without it, every design principle violation reaches merge — at AI speed.

Today — the incomplete pipeline
1
PR Opened
code pushed to branch
2
Build
compile · lint · type-check
3
Tests
unit · integration
4
Code Quality
SonarQube · ArchUnit
?
No design gate
violations pass silently here
5
Code Review
human PR review
6
Merge
code ships
With rkito — the complete pipeline
1
PR Opened
code pushed to branch
2
Build
compile · lint · type-check
3
Tests
unit · integration
4
Code Quality
SonarQube · ArchUnit
Design Audit
rkito CDA · automated · every PR
5
Code Review
human PR review

Reviewers see rkito's annotated comments inline — design intent context, violations flagged by CDA, and the architectural rationale behind each finding. No tribal knowledge required.

6
Merge
code ships
✓ Conformant → proceeds normally
⊗ Unplanned drift → blocked · CHI auto-drafted
→ Approved drift → proceeds with record
The Core Distinction
Code ReviewDesign Review

A PR can pass every SonarQube gate, every GitHub Copilot Code Review — and still be an architectural disaster.

rkito is the only platform that operates on design intent conformance. Not structural analysis. Not code quality. Design intent.

Code Review
Asks
Does this code pass?
Evidence
The diff.
Cannot catch
Design violations not
expressible as code rules.
Design Review
Asks
Does this decision conform?
Evidence
Diff + Architecture Ledger
+ Steering principles.
Produces
A durable architectural
record of approval.
Agentic Engineeringbefore the first commit · inside the agent session

Shift
Left.

DesignIntentrkitoSessionProtocolcontext loaded hereWritePushBuildLintTestsrkitoCDA

Every AI coding agent
is design-intent blind.

No conformance constraints. No architectural boundaries. No design intent. The agent generates code that compiles — and violates everything it cannot see.

The rkito Session Protocol is the mechanism by which agents operate precisely within design intent — retrieving the approved design, staying within CHI scope, and implementing exactly what was decided. Nothing less. Nothing more. The Skills Registry provides the governed instruction layer that makes this possible for any entity in the Architecture Ledger.

Today — context-blind agent
start
Agent session starts
no registration · no governance context
read
Reads codebase
no Steering · no Ledger state · no Lenses
?
No design context
Skills · Lenses · CHI · Ledger — none loaded
gen
Generates code
syntactically correct · intent-blind
PR
PR opened
CDA finds violations · code already written · rework required
Violations discovered too late. Repair cost is high.
With rkito MCP — context-first agent
auth
Session registered
register_session · tracked in Agent Shepherd
skill
Skills + Lens bundle loaded
get_skills() → instructions + applicable Lenses + Steering refs
CHI
CHI gate — intent approved
get_approved_chi() · no code before architectural intent is approved
ctx
Full context assembled
Ledger state · applicable Lenses · Steering context — all loaded
gen
Governed generation
validate_diff() at each write · BLOCKER findings halt immediately
PR
PR opened — already clean
CDA confirms what the agent already validated · no surprises
✓ Violations prevented — not caught
◈ Skills carry the Lens bundle — agent cannot be instructed to violate
⊕ Same governance record as CDA — no divergence
What shift left delivers
Beforefirst write
Design intent, Lenses, and architectural boundaries loaded before the agent touches a file
0surprises at PR
When validate_diff runs during generation, CDA finds nothing new at merge time
100%session trail
Every agent session tracked — auth, Skills loaded, CHI gate, each write, PR close
The Cost of Architectural Blindness

100 agent PRs.
What changes when they have context.

Composite
Teams shipping ≥50% AI-generated code
rkito MCP enabled vs disabled

Without rkito MCP
With rkito Session Protocol
ADRIFT
Agent-Introduced Drift Rate — undefined, untracked, unmeasured
≈ 0%
ADRIFT when agents load Skills and validate_diff before raising a PR
3–5×
Rework loops per blocked agent PR — agent writes, CDA blocks, architect reviews, agent rewrites
0
Rework loops — violations caught during generation, not after merge attempt
Unknown
Architectural scope of agent changes — no CHI, no declared intent, no approved boundary
Declared
Every structural agent change scoped and approved via CHI before the first commit
Invisible
Active agent sessions and what entities they are modifying, right now
Agent Shepherd
Live operational view of every active session — entities modified, Skills loaded, CHI status
The Cost of the Missing Gate

100 PRs.
Four numbers that change.

Composite sample
Teams shipping ≥30% AI-generated code · 2025
No design audit gate in pipeline

Without rkito
With rkito CDA
41
PRs needed multiple rework iterations
0
Design questions answered before code review opens
23
PRs stalled waiting for an architect
0
CDA runs in minutes — no architect required
12
PRs reverted after they shipped
0
Violations caught at the gate, before merge
4.7 days
Average PR-to-design-clean-merge
< 5 min
CDA runtime — automated, every PR, no exceptions
Research

Metrics that
didn't exist.

0tools produce a Design Conformance Score
0platforms measure Agent-Introduced Drift
0competitors have a Design Health Score

rkito introduced 35+ design fidelity metrics — architectural conformance, drift, and governance signals no code quality tool, CI pipeline, or architecture platform measures. This is what Design Continuity, Cognitive Debt, and Architectural Drift look like as numbers your org can trend, not just talk about.

Read the research
Open Source Orgs on rkito

Explore governance
built by the community.

OSS projects on rkito publish their lenses, policies and best practices openly. Browse them without an account — adopt them into your org once you sign up.

Create free account to adopt
Getting Started

Up and running
in an afternoon.

Four steps. No bespoke tooling. No onboarding engagement. The first design audit on a real PR — the same day you start.

Full setup guide
01
Adopt a Steering foundation

AWS Well-Architected, OWASP Top 10, ISO 27001 and more are pre-built inside rkito. One click adopts them into your org. You have design intent on day one — without writing a single artifact from scratch.

02
Point rkito at your repo

Connect your GitHub repository. rkito reads your codebase, Terraform, Helm, Kubernetes manifests and OpenAPI specs — then automatically populates your Architecture Ledger with every component, boundary and interface it finds.

03
Review and commit your baseline

The scan generates a Change Intention. Architects review it, update what needs updating, and merge. That is your living Architecture Ledger — the verified baseline every future PR is audited against.

04
Install the GitHub App

One install. The PR webhook is configured automatically — no YAML, no callback setup. The next pull request your team opens gets a full design audit before it reaches code review.

Done. CDA is live.

Every PR from this point is audited against your Steering standards and Architecture Ledger — automatically, before it reaches code review.

Already have design decisions documented?Coming soon

Import your existing Confluence pages, Word documents, ADR folders, and policy PDFs. rkito's onboarding AI reads them and converts them into structured Steering artifacts and Lenses automatically — so your entire architectural history is in the system from day one, not just what you write going forward.

Deployment

Every deployment
model. Every requirement.

Compliant, regulated, and air-gapped organisations have non-negotiable constraints. rkito meets them — without compromising the governance model.

Available now
GitHub Actions

Drop in as a single workflow step. No new infrastructure, no agents to manage. Running in your first PR the same afternoon.

Early access
rkito Cloud

Fully managed on AWS. Data residency compliance available — choose your region, your data stays there.

Enterprise
On-Premises

Docker container. Air-gapped. No data leaves your network — designed for regulated industries and government environments.

SOC 2 alignedGDPR · CCPAData residency controlsAir-gap capableNo training on your data
Open Source Program
Full platform.
Free forever.

Every qualifying open source project. All three layers. Unlimited members. The same platform paid teams use — no trial clock, no upgrade pressure.

Criterion 01
Public repo
+ OSS licence

Recognised open source licence. Public repository. That is the full qualification criteria — nothing else.

Criterion 02
Annual
re-validation

Self-service. No sales call. Stays free as long as the project qualifies. No upgrade clock.

Criterion 03
Powered by
rkito badge

Required. The way open source projects help build the category — and how we say thank you.

Apply to OSS ProgramSelf-service application — no sales conversation required.

Start with Starter. It's free.

Five members. The full three-layer platform. No credit card.

If you're on an OSS project, start there instead.

Signing up creates a Starter organization. Steering, Architecture Ledger, Design Reviews, ADRs, and GitHub integration are all included.