GVERA turns an idea into verified CAD, circuit boards, firmware, simulation results and launch material, on your own computer.
You describe what you want. GVERA interviews you, plans the work, then builds and checks each piece, from thirteen steps to a finished project. Every step ends at a gate that must pass before the next one starts, and every result is shown with its evidence.
Follow the spider desk-walker project, from its brief to its actual 3D assembly. Click a step to explore it in the app’s updated Simple view.
New project
A scripted replay based on the spider project, using its actual CAD assembly and concept sketches. The successful outcome, checks, timing, spend and file counters are illustrative.
Four phases. The pipeline interviews you, plans, designs, builds and tests, then ships. A phase is only done when its gates say so.
Intake, ideas, requirements, plan
GVERA asks the questions a good engineer would, then writes down what the product must do, which technologies it will use and what it may cost, weigh and draw.
Sketches, mechanical CAD, electronics
Four concept sketches that differ in architecture, not just color. You pick one, and GVERA designs parametric parts and the circuit board around it.
Software, simulation, validation
Firmware and software are built and tested for real. Physics simulation runs in MuJoCo, and a final pass traces every requirement to the evidence for it.
Instructions, launch material, handoff
Assembly instructions and a runbook, a product video in three aspect ratios, a landing page that only claims what the project can show, and a handoff note with what GVERA learned.
The app does not stop at a green tick. It shows which checks passed, which requirements are proven, and where a stand-in is doing the work.
A test project, a 7-inch inspection quadcopter, run with a scripted model so that all thirteen steps could be exercised end to end.
Each step has typed gates: deterministic checks, analysis, simulation, an independent reviewer model, or a human sign-off. A step that fails is repaired in place, and only the parts that failed are redone.
If a controller is a stand-in or a requirement has no proof yet, the page says so.
Every event is hash-chained. gvera journal verify detects tampering, and a killed run resumes from its last checkpoint.
Only the things that cost money to run live on our servers. Everything GVERA makes for you does not.
The app, the pipeline and every file it produces.
Accounts, your credits and the model gateway.
From $5one-time
Buy credits when you need them: $1 buys $1 of AI work, and every project you start draws on the same balance. Start as many projects as your credits cover. Checkout is handled by Dodo Payments from inside the app.
In a normal folder on your computer. GVERA reads and writes it directly, and every change is committed to Git. On macOS the first launch asks for access to the folder your projects live in. Allow it once.
Yes, for model calls, which go through GVERA Cloud, and for the first-run setup of the engines. The CAD, simulation and checking steps run on your machine.
The GVERA app and its private Python environment with CadQuery, trimesh and MuJoCo. gvera setup then adds Git, FFmpeg, KiCad, Node.js and a browser if they are missing. The Windows installer does all of it for your account only, without administrator rights for GVERA itself.
GVERA is not notarized by Apple, so macOS blocks the first open of a copy downloaded in a browser. Allow it once in System Settings, Privacy and Security, with Open Anyway. Or install it from Terminal with curl -fsSL https://gvera.app/install.sh | sh, which needs no approval. The Windows installer is unsigned too, so SmartScreen asks you to choose More info, then Run anyway. The download page walks through both.
Trust the evidence it shows, not the word verified. Gates prove that parts fit, boards pass their rule checks and tests pass. They do not replace a bench test of a physical build, and the app lists what is still unproven.
The session is persistent. GVERA checkpoints at every stage boundary and resumes after a crash, a kill -9 or a restart. A failing step is retried with feedback, escalated, and if it still cannot pass it blocks and tells you why, while independent work carries on.
