Understand the product
Build a durable model from source, routes, components, OpenAPI, approved traffic, requirements, tests, and the current change.
AI Agentic Verification Harness · local-first · evidence-backed · built by testers.ai
Testing that keeps up with the code.
CARBON turns the context inside your coding agent into a product map, risk-ranked tests, real execution evidence, and a regression system that learns with every build.
Illustrative scripted demonstration—not measured results.
›$carbon-test "checkout recovery"
01CHANGE DETECTED
Checkout recovery changed.
WHY THIS TEST?A failed payment crosses UI, API, and stored state.
MAPRISKPLANTESTVERIFY
ObservedWaiting for evidence
Confidence42 → 42
Top riskPayment recovery
EvidenceChange context only
Built into CARBONA reusable starting library for common components and product flows. Candidates become passes only after CARBON executes them and captures evidence.
3,745built-in candidate tests for common components
42built-in reusable product flows
44local MCP tools
54coding-agent workflows
One verification harness
From “test this” to evidence you can ship on.
CARBON coordinates the tools your coding agent already has. It preserves the difference between a generated idea, an observed result, and a regression you can trust.
Choose the right depth
Select breadth, depth, or balanced exploration from risk instead of running the same blunt suite after every change.
See the run before it starts
Open a protected live plan with exact tests, approval gates, and a time estimate—then watch status and remaining time update from observed execution.
Capture the whole page
Keep screenshots, DOM and accessibility state, console, network, performance, safe data, traces, and recordings wherever available.
Discover unfamiliar behavior
Map unknown widgets, APIs, states, transitions, and custom flows without silently converting hypotheses into expected behavior.
Make confidence visible
See confidence in quality, remaining risk, achieved and missing coverage, and what evidence would change the decision—alongside screenshots, persona feedback, exploration, and build comparisons.
CARBON Feedback Loop
Bring real reviewers into the build: scoped artifacts, reviewer teams, document comments, context checks, fixes and retests. Remote/ReviewLoop functionality is included, with a CARBON-branded server you can host privately or publicly on your own infrastructure. Sharing stays explicit.
Explore the feedback loopCARBON Browser v1
Run Vibium. Add the rest of CARBON.
CARBON can execute the official Vibium CLI and existing Vibium tests without rewriting them, then add generation, exploration, evidence, regressions, confidence analysis, and visual reporting through its own stable browser contract.
Portable intent
CARBON keeps what the test means separate from a brittle selector or vendor-specific command.
- Roles, accessible names, labels, text, test IDs and stable references
- Sessions, pages, discovery, actions, waits, assertions and recording
- Executable official Vibium CLI plus existing JavaScript and Python Vibium tests
- Capability contracts for host-agent, Playwright/CDP, Selenium/WebDriver and managed devices
Evidence by contract
Before execution, CARBON negotiates what the adapter can do and what evidence it can return. Missing capability is never silently hidden.
- Before/after state, actual result, timestamps and adapter version
- Screenshots, DOM, accessibility, console, network, performance, trace and video handles
- Redacted durable ledgers under
.carbon/browser/sessions/
Vibium compatible
Keep the browser API. Gain a complete verification harness.
Use familiar go, map, find, click, fill, wait, screenshot, and recording commands—or run an existing Vibium test file. CARBON preserves the native result in a durable evidence session.
Native compatibility
CARBON invokes the official local Vibium binary. It does not fork Vibium or silently install it. An npx fallback requires explicit opt-in.
carbon-runtime vibium exec --root . -- map carbon-runtime vibium run-test --root . --file tests/login.mjs
CARBON extensions
Generate project-aware tests from a Vibium map, manage prompt-defined intent tests, explore by breadth or depth, collect personas and accessibility evidence, promote regressions, and report remaining risk.
carbon-runtime vibium generate --root . \ --target https://your-site.example \ --request "Generate risky checkout tests"
Complete coverage
More than browser automation.
CARBON tests the product boundary: what users see, what services do, what AI systems decide, and what evidence would change the release decision.
Product testing
Generate from the real surface, then execute only the authorized checks that matter.
- Web pages, responsive states, accessibility and visual behavior
- APIs, schemas, auth, boundaries, idempotency and resilience
- TestRail, Xray and multi-format tests converted to AI intent
- Global credentials, reusable prompts and local context documents
- Full local bake-in with bounded test repair and evidence-gated activation
- Persona feedback and breadth/depth exploratory charters
- Incremental, scheduled and bounded test-fix-retest loops
AI confidence
Design evidence around non-determinism, populations, judges, safety, and release decisions.
- Prompts, models, RAG, ranking, agents and multimodal systems
- Repeated trials, slices, calibration and statistical review
- Prompt injection, data leakage, tool authority and side effects
- Independent eval-design, skeptical and release reviews
- Ship, canary, hold, rollback or request-more-evidence decisions
Plugin architecture
Thin commands. One CARBON runtime.
The harness implementation and test-pack knowledge live in the installed plugin runtime. Host skills stay small and route every supported coding agent to the same engine.
Pricing
Start with the harness. Add people when you need them.
Run CARBON yourself, add direct email support, work with us over Zoom on targeted customization and adaptation, or bring in experienced help to transform how your team approaches QA with AI.
01 / COMMUNITY
Open source soon
Free and soon open source
$0forever
Use CARBON locally and build repeatable test knowledge inside the coding agent you already use.
- CARBON plugin and local runtime
- Built-in component tests and product flows
- Evidence-rich reports and confidence analysis
- Self-service documentation
02 / SUPPORT
Monthly
Email support
$500/ month
Get direct help installing CARBON, shaping workflows, troubleshooting runs, and interpreting the evidence.
- Everything in Free
- Direct email support
- Workflow and setup guidance
- Help understanding results and next steps
03 / GUIDED
Monthly
Zoom support + customization
$2,500/ month
Work with us live to adapt CARBON to your product, workflows, and team, with targeted hands-on help each month.
- Everything in Email support
- Scheduled Zoom support
- CARBON setup and workflow adaptation
- Some customization and implementation work
04 / AI QA TRANSFORMATION
Team transformation
Transform QA with AI—confidently.
We help you rapidly transform your QA teams and processes with AI—confidently, based on proven experience.
$10,000/ month
- Analyze current test processes and artifacts
- Add AI to existing testing workflows
- Establish AI-first test coverage
- Help educate the team and improve adoption dynamics
Listed prices are in USD. Transformation and customization scope is agreed before work begins.
CARBON
Give your coding agent a verification harness.
Start locally with no testers.ai API key. Install the plugin, point your agent at the project, and ask CARBON Help to recommend the right first run.