# My ICP project: an encrypted backup application

**URL:** <https://forum.dfinity.org/t/my-icp-project-an-encrypted-backup-application/76153>\
**Category:** Showcase\
**Created:** [October 19, 2026, 2:36pm UTC](https://forum.dfinity.org/t/my-icp-project-an-encrypted-backup-application/76153 "2026-10-19T14:36:17Z")\
**Posts on this page:** 8\
**Page:** 1

<div class="post-metadata">

**Author:** ![ImSteffi](https://sea1.discourse-cdn.com/flex023/user_avatar/forum.dfinity.org/imsteffi/32/19792_2.png) [@ImSteffi](https://forum.dfinity.org/u/ImSteffi)\
**Post date:** [October 19, 2026, 2:36pm UTC](https://forum.dfinity.org/t/my-icp-project-an-encrypted-backup-application/76153/1 "2026-10-19T14:36:17Z")

</div>

Hello ICP community,

Over the past three years, I’ve been working toward a career in software engineering. I’ve spent more than a year learning about the Internet Computer, and I’d like to share a project I’ve been building: a browser-based encrypted backup application.

I’m a junior developer, and I started this project hoping to become a solo founder and build a backup solution that businesses could use. As I worked on it, I began to understand how much more it takes to turn an implementation into a reliable product that companies can trust with their data.

That has been a humbling learning experience. I’ve decided to share the idea and code publicly, get feedback, and learn from people with more experience. I’m still improving how I structure code and design systems, but I’m proud of how far I’ve come.

**Source code:** [https://github.com/ImSteffi/ICP\_backup](https://github.com/ImSteffi/ICP_backup)

**What the application does**

Users can select files and folders, encrypt them in the browser, distribute the encrypted chunks across multiple storage canisters, and later download and reconstruct the original files.

The current access model is a shared backup workspace: approved users have access to the same workspace key and backups.

**Architecture**

The frontend uses React and Vite. The backend and storage canisters are written in Motoko.

The current configuration includes:

- A frontend asset canister.
- A backend canister for accounts, permissions, backup manifests, and storage coordination.
- Five storage canisters that store encrypted chunks using Motoko’s persistent Region storage.

The browser uploads encrypted file chunks directly to the storage canisters. An encrypted manifest records the file paths, chunk order, canister IDs, and storage offsets needed for restoration.

One design goal is to widen distribution across additional subnets by adding storage canisters. Upload distribution already uses the number of available canisters, and restoration follows the locations recorded in each manifest. Expanding beyond the current five still requires changes to registration, authorization, and chunk sizing, followed by end-to-end testing. Existing backups are not automatically redistributed.

The storage distribution is striping: different canisters hold different chunks. I have not implemented additional application-level replication or erasure coding.

**Encryption and access**

Files and manifests are encrypted in the browser using AES-256-GCM, with a fresh random IV for each encryption operation.

Authentication uses Internet Identity. There are three account roles: requester, user, and admin. An administrator approves access, and the frontend uses vetKeys-based identity-based encryption to wrap the shared backup key for the approved user’s principal.

Key rotation and access revocation are areas I still need to improve. Removing a user’s role restricts further API access, but it does not rotate the shared key or revoke data they have already downloaded.

**What I’ve implemented**

- File and folder selection, including multiple backup sources.
- Client-side encryption and encrypted backup manifests.
- Chunk distribution across five storage canisters.
- Parallel upload lanes, with worker-based encryption where supported.
- Content-based deduplication within an upload.
- Downloading, decrypting, and reconstructing files and folder paths.
- Account approval and role management.
- Logical backup deletion.
- An initial storage compaction mechanism intended to reclaim space occupied by deleted backups for reuse. It relocates active chunks and updates their manifest offsets. The implementation still needs fixes to prevent overwriting unread chunks and to recover safely from interruptions; it does not shrink already allocated stable memory.

**Current stage**

This is a development project. It has not been independently audited, and I’m not presenting it as a production-ready backup service.

Storage compaction needs corrections to its data-movement and interruption-recovery logic. Provisioning additional storage canisters, managing the shared key lifecycle, and recovering from interrupted operations also need more work and testing.

I’d also like to measure upload and restore performance, cycle costs, and browser memory use with larger datasets.

**Feedback I’d appreciate**

- Does a shared encrypted backup workspace address a useful need in the ICP ecosystem?
- What would you change about the storage and manifest architecture?
- How would you approach key rotation and access revocation?
- Which failure and recovery scenarios would you test first?
- How could I make the code easier to understand, maintain, and test?

Building this has given me practical experience connecting a frontend, authentication, browser cryptography, Motoko services, and persistent storage into one application. It has also helped me understand how much of a backup system’s design comes down to handling failures and making restoration dependable.

I’m looking for opportunities to contribute to a team and continue learning, including junior software engineering roles or project work involving ICP, Motoko, or frontend development. If this project is relevant to your work, I’d welcome a conversation.

Thanks for taking a look. I’m proud to share this milestone, and I’d appreciate honest technical feedback.

---

<div class="post-metadata">

**Author:** ![whitescreendev](https://sea1.discourse-cdn.com/flex023/user_avatar/forum.dfinity.org/whitescreendev/32/40650_2.png) [@whitescreendev](https://forum.dfinity.org/u/whitescreendev)\
**Post date:** [October 20, 2026, 12:23pm UTC](https://forum.dfinity.org/t/my-icp-project-an-encrypted-backup-application/76153/2 "2026-10-20T12:23:51Z")

</div>

The part I’d be most curious to test is what happens during a restore if one of the storage canisters is temporarily unavailable. Since the chunks are striped without extra redundancy, I wonder how clearly the app would report which data couldn’t be recovered. That seems like a useful failure case to explore early on.

---

<div class="post-metadata">

**Author:** ![AbelOdoh](https://avatars.discourse-cdn.com/v4/letter/a/7c8e57/32.png) [@AbelOdoh](https://forum.dfinity.org/u/AbelOdoh)\
**Post date:** [October 21, 2026, 11:31pm UTC](https://forum.dfinity.org/t/my-icp-project-an-encrypted-backup-application/76153/3 "2026-10-21T23:31:28Z")

</div>

Hi @ImSteffi, really nice work, especially encrypting everything in the browser before anything reaches the network, and using vetKeys to share access. Splitting chunks across five storage canisters is a neat touch too.

I help run a well-known ICP media account on X, and we’re starting a series highlighting things people are building on ICP. Would you be OK with us featuring your project? We’d describe it as an early development project, as you’ve said, not a finished service. If you’re on X, share your handle and we’ll tag you.

One question: how long does a typical backup or restore take right now, with the chunks spread across five canisters?

---

<div class="post-metadata">

**Author:** ![ImSteffi](https://sea1.discourse-cdn.com/flex023/user_avatar/forum.dfinity.org/imsteffi/32/19792_2.png) [@ImSteffi](https://forum.dfinity.org/u/ImSteffi)\
**Post date:** [October 25, 2026, 5:11pm UTC](https://forum.dfinity.org/t/my-icp-project-an-encrypted-backup-application/76153/4 "2026-10-25T17:11:03Z")

</div>

That’s a useful failure case to test, thank you. I haven’t tested it yet. With the current striping design, files that depend on chunks from an unavailable canister couldn’t be fully restored until it becomes available again.

The app should clearly distinguish between a temporary retrieval failure and permanently lost data, identify which files are affected, and allow retries. I need to improve and test that behavior.

I chose striping partly to keep storage costs down, but it comes with this trade-off. Replication or erasure coding would be worth exploring for stronger recovery guarantees. I’d be interested to hear how others would balance that against the cost to the end user.

---

<div class="post-metadata">

**Author:** ![ImSteffi](https://sea1.discourse-cdn.com/flex023/user_avatar/forum.dfinity.org/imsteffi/32/19792_2.png) [@ImSteffi](https://forum.dfinity.org/u/ImSteffi)\
**Post date:** [October 25, 2026, 5:21pm UTC](https://forum.dfinity.org/t/my-icp-project-an-encrypted-backup-application/76153/5 "2026-10-25T17:21:47Z")

</div>

Hi, thank you for offering to feature the project! Yes, I’d be happy for you to include it, and I appreciate you describing it as an early development project.

I’m not active on X, but my handle is [@Hi\_ImSteffi](https://x.com/Hi_ImSteffi), so feel free to tag me. You can also find me on [LinkedIn](https://www.linkedin.com/in/steven-rekand/).

For upload performance, from what I remember of my informal tests, individual chunk uploads generally completed in around 1-3 seconds. With five parallel upload lanes and roughly 2 MB per chunk, I was seeing approximately 10 MB uploaded every 2-3 seconds, or around 3-5 MB/s.

Please treat those as rough observations rather than verified benchmarks. I still need to measure total backup time, including browser-side processing and manifest publication, across different datasets. I don’t have a reliable restore-time measurement to share yet.

Thanks again for helping bring attention to the project!

---

<div class="post-metadata">

**Author:** ![Lorimer](https://sea1.discourse-cdn.com/flex023/user_avatar/forum.dfinity.org/lorimer/32/17601_2.png) [@Lorimer](https://forum.dfinity.org/u/Lorimer)\
**Post date:** [October 25, 2026, 5:45pm UTC](https://forum.dfinity.org/t/my-icp-project-an-encrypted-backup-application/76153/6 "2026-10-25T17:45:30Z")

</div>

Why does this need to run on a blockchain?

Someone can locally encrypt a file and store it remotely with whatever redundancy they need (using multiple different services if they want, including local storage), without requiring something like ICP.

So my question is, what’s the problem statement, and then how is that addressed by this product?

---

<div class="post-metadata">

**Author:** ![ImSteffi](https://sea1.discourse-cdn.com/flex023/user_avatar/forum.dfinity.org/imsteffi/32/19792_2.png) [@ImSteffi](https://forum.dfinity.org/u/ImSteffi)\
**Post date:** [October 25, 2026, 6:15pm UTC](https://forum.dfinity.org/t/my-icp-project-an-encrypted-backup-application/76153/7 "2026-10-25T18:15:52Z")

</div>

My original idea was to back up a shared folder on a company’s RAID-backed storage, automatically uploading changes as users modified files.

My assumption was that companies backed this up to a centralized cloud, or multiple providers. If a provider became unavailable, the copy stored there would be inaccessible. That led me to ICP: a canister’s dataset is replicated across its subnet’s nodes, so availability doesn’t depend on one server. I saw adding more subnets as a way to spread storage and potentially increase parallel upload throughput, although I haven’t validated that scaling.

I understand that conventional cloud services also replicate data, so that alone doesn’t establish an advantage. As a junior developer, I got ahead of myself building a solution before validating the need. I’m honestly no longer sure whether this fills a gap that existing backup solutions don’t already address.

---

<div class="post-metadata">

**Author:** ![Lorimer](https://sea1.discourse-cdn.com/flex023/user_avatar/forum.dfinity.org/lorimer/32/17601_2.png) [@Lorimer](https://forum.dfinity.org/u/Lorimer)\
**Post date:** [October 25, 2026, 6:36pm UTC](https://forum.dfinity.org/t/my-icp-project-an-encrypted-backup-application/76153/8 "2026-10-25T18:36:32Z")

</div>

I think the uptime / availability / liveness aspect would be a potential advantage over traditional systems. But I’m not clear on where the use case exists for that (cloud platforms have SLAs and if they go down across the board is usually only briefly and very rarely). Users can also make use of more than one cloud provider to get additional redundancy - this of course can be automated and built into the service, which would be the sort of thing you’d be providing - however you’d want disk storage rather than in-memory, and you’d want something like erasure coding to provide efficient redundancy.

ICP’s blob storage was meant to be provide this (and will do whenever it’s finished, but it appears to have now been deprioritised by DFINITY for other things).

I’m sceptical that another service layer is actually needed for this sort of thing. I recommend starting with clear problem statement(s), where a strong advantage is offered by the on-chain aspect.

Out of interest, what are you interested in? If AI appeals to you at all you could start exploring the early version of the Internet Intelligence Gateway, and building around that (make sure to have a clear problem statement though).
