Open AI engineering environment

Build software.
Keep control.

Cude turns an idea into a visible engineering run—requirements, decisions, code, previews, and proof. Choose the model. Inspect the work. Own every file.

Apache-2.0 · Local-first · Provider-independent
CUDE / PRODUCT RUN
00:42
BUILD · STEP 03

Ship the working surface

RUNNING
Create an accessible project dashboard. Preserve the existing data model and show every changed file before applying.
app/routes/dashboard.tsx+142
1export function Dashboard() {
2  const run = useCudeRun();
3  return <Workspace
4    files={run.files}
5    evidence="required"
6  />;
7}
12 checks passedReview changes →
One environmentFrom intent to verified code
ModelsHosted or local
OutputReal, editable files
LicenseApache-2.0

One run. Every decision visible.

Cude structures the messy middle between a prompt and a product. You can follow the work, change direction, and verify what was actually built.

01

Brief

Define the outcome and constraints.

02

Architecture

Make technical choices explicit.

03 / ACTIVE

Build

Create in a real, inspectable workspace.

04

Preview

See the result while it is made.

05

Evidence

Finish with tests, diffs, and proof.

Your model is a choice.
Not a lock-in.

Connect a hosted provider, an OpenAI-compatible endpoint, or a local Ollama model. Switch as the work changes without giving up the workspace.

OAOpenAI-compatibleYour endpoint, your keyREADY
OLOllamaRun models on your machineLOCAL
ANAnthropicConnect when the task needs itOPTIONAL
+More providersOne workspace, multiple routesEXPAND

Not just an answer. A place to build.

Plan, edit, run, inspect, and hand off without losing the thread between product intent and implementation.

Files stay first-class.

Browse the project, inspect diffs, and keep generated work editable from the first line to the last.

product-brief.md   ·   architecture.md
# Product outcome

Build a fast, accessible workspace
that keeps every decision reviewable.

## Constraints
— preserve the repository
— use the selected model
— require evidence before handoff

Continue what exists.

Import a repository, respect its structure, and make the smallest safe change.

+12 reviewable files

Web, desktop, mobile, extensions.

Shape the run around the product—not the other way around.

The work remains yours.

Cude is licensed under Apache-2.0. The source is inspectable, the files are portable, and the provider boundary stays replaceable.

What you keep

  • Source fileseditable
  • Architecture recordvisible
  • Git historyportable
  • Verification evidencereviewable

What you choose

  • AI providerreplaceable
  • Deployment targetindependent
  • Project structurepreserved
  • When to apply changesyour call

Make the work
make sense.

Open Cude, bring a repository or begin from an idea, and leave with software you can explain.

Explore Cude on GitHub ↗