Open source workflow memory engine

Your computer should remember what you taught it.

Marouba learns your workflows once, replays them forever, and repairs itself when things change. Teach your computer how work gets done. Once.

[Marouba] Loading workflow: comfyui-generate-image
[Marouba] Route order: api, cli, uia, visual, ask
[Marouba] Trying route: api
[Marouba] Prompt queued: 8b31...
[Marouba] Complete. Output: vault/runs/output.png
[Marouba] Run logged.

vault/workflows/comfyui-generate-image.md
status: portable, inspectable, yours

The problem

The frontier labs are building smarter agents. None of them are building ones that remember.

Every AI agent today starts from zero. Every session. Every time. You teach your computer to export a report, process an image, or run a production task. Tomorrow it forgets. The industry's answer is bigger models and longer context windows. But the problem is not intelligence. It is memory.

Every workflow is re-taught

The average knowledge worker repeats the same digital workflows dozens of times per month.

Every update breaks brittle automation

A button moves, a dialog changes, and pixel-click loops are completely lost.

Every fix disappears

When a route breaks, the correction vanishes into the session instead of becoming shared knowledge.

The solution

Marouba: teach once. Replay forever. Repair itself.

Teach

Perform your workflow normally. Marouba observes the context, the route, and the intent behind what you did.

Replay

Need the same workflow again? Marouba replays it from the vault, even when the screen layout changes.

Repair

When something moves or breaks, Marouba asks once, writes the fix back, and never asks again for that failure.

See how it works

A 60-second demo flow.

This is the placeholder for the launch video: a real desktop workflow taught once, replayed the next day, then repaired after an app update.

00:00 This is my computer.

A quiet desktop. The same repetitive workflow is about to start.

00:08 Today I am teaching it once.

Open Photoshop, apply the same filters, export, save. Marouba records the intent, not just the clicks.

00:15 Tomorrow, I hit replay.

The workflow runs from the vault. No re-teaching. It just knows.

00:21 Then the app changes.

Photoshop updates. A button moves. This is where every other automation breaks.

00:28 Marouba repairs the workflow.

It finds the new path, completes the task, and writes the repair back to the vault.

00:49 Teach once. Replay forever. Repair itself.

The same loop works across Photoshop, Excel, Ableton, Chrome, and any desktop app with a profile.

The vault

Not a macro. Not a script. A living record of how you work.

Traditional automations store actions: click here, type this, press Enter. Marouba vaults store intent, routes, fallback paths, screen state, verification rules, and repair history. They are git-friendly, human-readable, portable, inspectable, and yours.

---
id: photoshop-export-for-web
app: Photoshop
routes:
  - type: cli
    command: "photoshop.exe -r export.jsx"
  - type: uia
    element: Save for Web
  - type: shortcut
    keys: [ctrl, alt, shift, s]
fallback_order: [cli, uia, shortcut, ask]
verification:
  type: file_exists
  path: "{output_path}"
---

Marketplace

Your workflows have value. You should be able to share it.

Built a perfect Photoshop export workflow? A bulletproof Ableton mastering chain? A reliable ComfyUI production route? Publish it as a signed vault. Buyers download it, it adapts to their machine, and every repair makes the workflow more trustworthy.

Creator 70%
Platform 30%

Waitlist

Stop teaching your computer the same thing twice.

Early members get first access to the beta, input on the vault format, founder status in the marketplace, and a direct line to the core team.