Last updated: August 2026
We're a small team, so we won't claim certifications we don't have. Here's a straightforward account of what we actually do.
Official iCoderAgent releases are code-signed with an Apple Developer ID certificate and submitted to Apple's notarization service before being published for download — the same trust chain macOS's Gatekeeper expects from any legitimate direct-download app, even outside the Mac App Store. Verify it yourself after downloading:
spctl -a -vv /Applications/iCoderAgent.app shasum -a 256 ~/Downloads/iCoderAgent.dmg # compare with the checksum below
Current release checksum (SHA-256): see downloads/iCoderAgent.dmg.sha256. We also publish /.well-known/security.txt for vulnerability reporting.
The website and license-verification service are only served over HTTPS/TLS (via Amazon CloudFront and API Gateway). There's no unencrypted path to either.
We never see or store your card details. Payments are handled entirely by our payment processor, which is itself PCI-DSS compliant.
License and billing data live in a database with access scoped to only the specific operations our backend needs — nothing broader. Webhook events from our payment processor are cryptographically signature-verified before we act on them, so a request can't forge a license without a valid signature from that processor.
As covered in more detail on our Data Use page: your source code, prompts, and AI provider API keys are never transmitted to or stored on our servers. That's not a policy promise requiring you to trust us — it's a consequence of how the app is built (requests go directly from your Mac to your chosen AI provider).
The agent runs a guardrail layer on every message, on every platform. Inputs are checked for prompt-injection/jailbreak patterns and unsafe content, and the agent is hard-scoped to software development — off-topic sessions are refused before any work starts. Personal data and secrets (emails, phone numbers, card numbers, API keys and tokens) are detected and replaced with placeholders before they reach the model, and the agent's replies are scrubbed the same way on the way back. These checks are deterministic code in the app, not a policy — and they're covered by an automated test suite that ships with every release.
Every action the agent can take is permission-tiered. Destructive operations (deleting files) always require your explicit approval — there is no configuration that bypasses it. File access is confined to the project workspace by a path-escape guard, and in planning mode all write/run/delete tools are hard-blocked. If the agent proposes a new tool for itself, you see the full source code and approve it before it can ever run.
Companion-pairing accounts use salted PBKDF2 password hashing (200,000 iterations), HMAC-signed session tokens with expiry, and constant-time credential comparison. Sign-in endpoints are rate-limited against brute-force attempts. Password-reset and verification tokens are single-use, expiring, and stored only as hashes. The account service stores credentials and usage counters only — never your code, prompts, or chat history.
We don't yet have a self-serve "delete my account" button — email support@aidigiapps.com and we'll delete your license/billing records on request, subject to what we're legally required to retain (e.g. tax records).
Found a security issue in the app, the website, or our backend? Email support@aidigiapps.com with details — we'll acknowledge reports promptly and let you know once it's addressed. Please don't publicly disclose an issue before we've had a reasonable chance to fix it.