Your photos, your AWS account — Argentic is on AWS Builder Center

Like a lot of people, I was paying a monthly subscription for photo storage — while being, professionally and personally, an AWS customer with an account that could store the same photos for a fraction of the price.

So I built Argentic: an iOS app that backs up your photo library to a bucket you own, makes it searchable by meaning (“sunset on the beach”), and gives you an AI assistant that can find, describe, and propose cleanup of your photos.

The full architecture story — with the real architecture decision records from the build — is now published on AWS Builder Center:

→ Your photos, your AWS account: building an agentic-AI photo backup for iOS with Kiro — no servers

Here are the three ideas that make it worth your click.

1. There is no server anywhere

No Lambda, no API Gateway, no container. The iPhone talks to S3, Cognito, and Bedrock directly.

The auth chain does all the heavy lifting: Sign in with Apple → Cognito Identity Pool → temporary STS credentials on the device. From there, the app presigns its own uploads, queries its own vector index, and invokes its own model — every call made with credentials that expire within the hour and that IAM confines to the signed-in user’s data.

The assistant itself is a plain Converse tool-calling loop running on the phone, with guardrails treated as architecture rather than afterthought: IAM allows exactly one model ID, tool rounds and tokens are bounded, and a CloudWatch alarm emails the account owner if usage ever spikes. The agent proposes deletions; only a human confirms.

2. Multi-tenancy in one line of IAM

Family sharing is enforced by AWS, not by app code:

arn:aws:s3:::<bucket>/photos/${cognito-identity.amazonaws.com:sub}/*

Each Apple ID maps to a stable Cognito identity; each identity is confined to its own prefix. A tenant cannot read — or delete — another tenant’s photos even with a raw S3 client. That is the whole tenancy model.

3. Semantic search without a per-photo AI bill

Captioning every photo with a cloud model would mean a per-photo cost forever, and photos leaving the account. Instead, everything per-photo runs on-device: Apple Vision for labels and OCR, a CLIP-family model (MobileCLIP-S0, CoreML) for embeddings, and S3 Vectors — in your account — as the index. Queries come back in ~100ms. Bedrock only ever sees the single thumbnail you explicitly ask the assistant to look at.

The result, for a typical 50 GB library on S3 Intelligent-Tiering: ~$0.35/month, total.

Built with Kiro, decisions on the record

The second story inside the article is the workflow: the app was built end-to-end with Kiro as the engineering agent — cross-session project memory, ADRs written before the code, verification gates on every change, and an honest division of labor where the human owns product pivots and destructive actions while the agent owns everything reversible.

The actual decision records from those sessions are published with the project — including the sequencing trap in the multi-tenant migration that would have bricked the installed app if IAM had been tightened before the app shipped.

Try it

• Read the full article on AWS Builder Center
• Join the beta on TestFlight (link in the article)
• The repo is finishing its open-source review and will be public very shortly at github.com/agiusalexandre/argentic — the decision trail (docs/adr) ships with it

Your photos never leave your account; the only AI that sees a photo is the one you explicitly point at it.