Test while your AIkeeps building.
Testing comes first. Experience, inspect, challenge, refine and approve AI-built software in task-specific testing environments. Precision tools and structured feedback put you in control of the result.
Our ambition: the most capable testing environment imaginable.
Generating code is notthe same as finishingsoftware.
How it looks. How it behaves. How it feels in someone's hands. If your name goes on it, almost right is not right.
Not acceptable.Not almost.Not after launch.
AI can generate the code. You bring the judgment. Bedz gives you the instruments to act on it.
Test while your AIkeeps building.
The old loop can fail before you even get to test. With Bedz, testable tasks arrive first in running Beds, so you can test and respond while your AI keeps building.
Slower on the left, faster on the right. One quarter of the original speed by default. Changes playback only, not the estimated times. Pause freezes both the timeline and viewing angle.Old build 1 · Planning. Bedz is planning tasks 1–10.
* Illustrative estimates · see every step
The five old builds have 2, 1, 4, 0 and 3 failed-run cycles. Each repair includes another download, extraction and attempt to run. Animation is accelerated. Every rectangle uses the same linear time scale: five minutes is ten times as long as thirty seconds. Labels repeat inside long rectangles.
| Old-way step | Each time | Human time |
|---|
| Ten-task build | Failed runs | Old elapsed | Bedz elapsed |
|---|
Bedz assumption: five minutes of planning per ten tasks, then a 12½-minute AI build. Five test-and-feedback passes take 2m 15s each, beginning after the first task arrives. Both methods budget the same testing, feedback and planning effort. The Bedz example assumes ready-to-test delivery; its error correction and retest time are not modeled. These are scenario estimates, not measured results or a performance guarantee.
Five old-way builds contain two, one, four, zero and three failures before testing. The lower lanes keep delivering numbered tasks across that entire clock, with planning before every group of ten. Each group delivers five testable tasks before five non-testable tasks, with five human test passes that begin after arrival.
Railz.
Test it running. Your real app, live, inside the rails. Record a run, snapshot a moment, flag a problem the second you see it.
Railz CT.
Refine it, in words. Mark exactly what falls short and say what you want. BedzPost hands it to your AI as an exact instruction, not another paragraph to guess at.
Frame.
Verify what comes back. Every task your AI finishes opens in its own Frame. Press GO, watch each check pass, then Accept. Not before. Or keep your hands on the keys: Ctrl+Enter runs GO, every check and Accept.
Frame CT.
Refine it, by hand. Drag it into place, read the exact numbers, and send the fix back to your AI, measurements and all.
All four.
One instrument, four positions. Top: your live app. Bottom: what your AI sent back. Left: look and test. Right: change it, with Craftsmanship Tools.
Action Frame.
The tools come to the work. Select anything in Railz CT or Frame CT and its tools appear around it: move, fill, resize, exact numbers. Your mouse never has to leave it.
Speed without caremakes unfinished workfaster.
Bedz isn't here to lower your standards so you can finish sooner. It's here to help you reach them sooner.
Not everyonewill notice.You will.
A button two pixels off. A transition that doesn't feel right. A feature that technically passes but isn't finished. You don't need to be a software engineer. You need to care like a professional.
Build with AI.Not behind it.
Quality should be demonstrated, not declared. Speed should remove wasted effort, not necessary care. And the person who stands behind the software decides what gets accepted.
- Bring the AI plan you already pay for. Bedz works with your own paid Claude or ChatGPT account. The Claude desktop app and ChatGPT hand their work to Bedz right on your computer.
- Out of usage? Keep building. Hit your Codex or coding-agent limit? A paid ChatGPT plan with GitHub connected can keep delivering work to Bedz, so the chat plan you already pay for becomes a second engine.
- GitHub is optional. A spare lane, never a requirement. Bedz never resells AI: you use your own plan, so there are no Bedz credits and no surprise bills.
What you buildis yours.
Your software lives on your computer. Real project files, in your own workspace. What you have built stays built when you change the tools you use.
Your files. Keep the work. Open it, change it, back it up or hand it to another developer.
Your AI. Bring your own paid plan. Work with supported providers and workers, with the same human acceptance workflow.
Your next move. Choose where your software runs and where it goes next.
Finished with care.Marked with proof.
When the work is complete, save the Bedz Hallmark. A shareable certificate of the changes you accepted, who approved them, and which checks passed.
A record of acceptance. The project, accepted changes, approvers and dates.
The checks behind the work. Recorded results and whether each change is live in the project.
Proof you can take with you. One page that opens in any browser, with its own data and fingerprint.
Nine qualifiers.One complete workflow.
Master Craftsmanship Vibe Coding Systems
Independent Testing Environments
Software must be executable and testable in task-specific isolation without disrupting approved project work.
Guardian Lane;
the Lane Guardian
Guardian Lane helps you keep track of the job, check what changed and pick up where you left off. Ask for help in everyday words.
“Check the result.” “Save a copy.” “Help me hand this over.” His reports explain what happened, what still needs checking and what to do next.
Six tools. Six useful jobs.
- Drift Guard
- Notice when changed files stray outside the job you agreed on.
- Zip Stripper
- Put useful project files in one ZIP to share, leaving out common clutter and private files. Check it before sharing.
- Handoff
- Give another AI the job, your decisions and selected files so it can continue.
- Proof
- Try named checks and read the results in plain words. Anything that needs your judgment stays marked for you.
- Checkpoint
- Save a copy of your files. See what would change before choosing to go back.
- Doctor
- Look for setup problems and get a clear next step.
These six tools are free and open source. Use them on their own, or through Guardian Lane in a Bedz build with the tools connection. Drift Guard warns you; it does not stop other programs. You decide what to keep.
Explore the Bedz featuresYour project.
Your way of working.
Build, test, refine and approve. Shape the tools you use along the way.
Eight ways to make the work yours. Open a feature to explore it.
01PC and MacA desktop workspace on your computer.
Windows PC: Bedz desktop beta. Mac: preview builds for Apple Silicon and Intel.
Early access is by invitation. Mac previews have passed native build and launch checks; full fresh-install and interaction testing, Apple signing and notarization are still in progress. Preview builds may differ from the current Windows tools.
Request early access02Add to FramezBuild tools for the way you test.
Ask your AI for an extra tool in your task-review workspace. Describe what it should do and where you want it.
A Frame holds the proposed work for a task. Your added tools can help you examine, adjust or check that work.
03Add to RailzExtend the workspace around your live app.
Request a tool that fits your project and the work you repeat. Railz brings the running application and your working controls together.
Use the supported My tools locations to add capabilities with your AI.
04Add to the Action FramePut a new tool where you need it.
Select the work, then use the tools around it. Move, resize, change appearance and enter exact values without searching through a distant panel.
Ask your AI to add a supported custom tool in the Action Frame location you choose.
05Test your tools firstTry the addition. Decide whether to keep it.
Your custom-tool request goes through BedzPost. The AI returns a candidate Frame for testing and acceptance.
Check that the tool does the job you asked for before accepting it into your workspace.
06Racing LinesChoose the task. See your next move.
Bedz opens the right screen, selects the relevant part and highlights the next real action. The guide advances when the work is completed, keeping required checks and the final choice with you.
One uninterrupted screen visit is one Racing Line. Several actions on that screen still count as one. Pause, explore and return; a correction records the extra visits.
Guided paths are in the separate Workbench preview. The public Community Lines catalog is live.
Explore Community Lines07Test your wayA grouped task list, or one TestBed per task.
Work through a grouped task list within a Bed to examine a connected experience. Use individual TestBedz to focus on separate tasks and proposed changes.
For example: check sign-up, sign-in and saving a profile as one journey, then use a task-specific Bed to examine a particular correction.
Check the complete journey as well as its individual parts.
08Build and test togetherYour AI keeps building while you check.
Receive testable tasks as they finish. Begin testing one while your AI continues with the next task or background work.
A Pass Pack names the passes, tasks and expected results, so both of you can see what is ready and what comes next.
See the concurrent workflowGo further.
The tools behind the workflowThe build-and-test loop
- A Frame for each task
- Open a named testable task with its proposed result and review context.
- Pass Packs
- Organize the project into passes and tasks with a clear “Works when…” expectation.
- Test and background tasks
- Separate work you can try directly from work that needs other checks. Deliver testable work early where practical.
- Separate candidates
- Examine proposed work in a separate environment while preserving the accepted project.
- Checks before acceptance
- Run the required checks and inspect their recorded results before keeping a change.
- Accept, change or steer
- Keep the result, request a correction or change direction. You decide what is accepted.
- Before and after
- Compare the proposal with the accepted version to see what actually changed.
- Feedback with context
- Keep the task, selected part, relevant version and evidence attached to the correction.
The craftsmanship tools
- Railz
- Try the real application running and flag a problem when you find it.
- Railz CT
- Refine the project with Craftsmanship Tools and precise instructions.
- Frame CT
- Adjust incoming proposed work before accepting it.
- Precision controls
- Drag into place, then use exact values, grids and rulers for finer adjustments.
- BedzPost
- Send the actual selected correction, measurements and instructions back to your AI.
- Locks
- Mark settled decisions for protection as further work continues. Protection follows the supported project rules.
- Freestyle
- Create and refine directly with supported page, 2D, 3D and motion tools. Available tools depend on the build.
- Your workspace
- Choose tool visibility and folded groups, adjust preferences and use keyboard controls.
Ownership, evidence and continuity
- Your supported AI account
- Bring your own supported AI plan and connection. Bedz does not resell AI credits; your provider’s limits still apply.
- Your project files
- Keep real files on your computer. Open, back up, change or hand them to another developer.
- Continue with another AI
- Keep the project and accepted work when moving to another supported AI connection, carrying the current plan and build with you.
- Record what happened
- Capture a run or a moment and attach useful evidence to a problem.
- Beyond the screen
- Use task-specific checks for calculations, saved data and background behavior that appearance alone cannot establish.
- History and recovery
- Inspect earlier work and use the available recovery controls when a change needs attention.
- The Bedz Hallmark
- Take away a record of accepted changes, approvers, dates and recorded checks. See the Hallmark ↗
- Community Racing Lines
- Browse and contribute supported action sequences. Only the line you choose to share goes to the catalog; your project files and personal visit history stay local. Browse the catalog ↗
New in the current source build
- Built · Tested · Approved
- See three progress bars against the same agreed outcomes, with evidence and a suggested next action. These source-build tools are awaiting release integration.
- Bring an existing project
- Start with a draft assessment Pass Pack. You and your AI review the feature inventory, credit evidenced existing work and identify missing checks.
- Test existing features
- Open a real test Frame for a supported browser feature. Its expected behavior and project version travel with the test.
- Ready to release
- Use a separate checklist for installation, recovery and handover. Full progress bars alone do not establish release readiness.
Beta and preview access is by invitation. Tools vary by build; Racing Lines guidance and the newest progress tools are identified above.
Help shape BedzPay once.Keep it.
Bedz Pro is a one-time purchase for the first 50,000 licenses, in four founding tiers. Early supporters pay the least, and every founding buyer keeps their terms for good.
Help shape Bedz.Be recognized for it.
Bring a real project and honest feedback. Complete an agreed testing assignment to qualify for the founding-tester rewards. Good feedback can be a test that fails.
- 12 months of paid-tier access.One year of the planned paid Bedz tier for qualifying participants. Your own AI usage is separate.
- 5 lifetime awards.Five personal lifetime licenses reserved for exceptional, substantive contributions. Criteria and full terms come before testing starts.
- Your founding identity.A numbered founding-tester designation and a permanent tester ID. Optional public credit for your contribution.
- An early voice.Share feature ideas and feedback. At 18 contribution points, unlock formal capability proposals for review.
- A place in the workshop.A private tester community for questions, help, ideas, announcements and following the work together.
- Recognition that stays.One contribution point per distinct completed recorded task after review, a personal thank-you and a lasting contribution history.
- 24 collectible badges.Bronze, silver, gold and platinum. Your first six credited tasks each earn a badge; further awards recognize milestones, failures, release coverage and retests.
- A level earned by doing.Progress from New tester to Explorer, Contributor and Pioneer. Your progress page keeps your points, badges and next milestone together.
The badge collection & how rewards work
Testing journey: First recording · Finding your feet · Pathfinder · Four corners · Building momentum · Steady contributor · In the flow · Double digits · Deep explorer · On a roll · Bedz pioneer · Full circuit · Long haul · Fifty strong · Bedz veteran · Centurion.
Reporting failures: Signal finder · Troubleshooter · Fault tracer. Across releases: Across releases · Release navigator. Following fixes: Back to the test · Follow through · Fix witness.
Completed recorded Pass and Fail tests receive equal credit after review. The same assignment counts once. Incomplete, blocked, cancelled, unconfirmed or unreviewed sessions do not earn points. Community recognition and license awards have their own qualification rules.
Testing is free to join. Access and places are confirmed by invitation; signing up does not guarantee a place or immediate access. Names and founding numbers are confirmed separately. These are beta tester benefits; paid founding-license purchases have separate terms.