Privacy and Data Handling
Last Updated: November 2026 • Architecture by SBF Consultancy (Pvt.) Ltd. (sbf-consultancy.net)
Project Pak-LLM Coder is committed to developer privacy. When using AI inference features, code context is automatically scrubbed by our client-side and edge Prompt Security Sentinel to redact high-entropy API keys, passwords, and private tokens before requests reach inference routers. Workspace auto-resume stores only your user-scoped active workspace pointer with zero cross-user data leakage, and our Android/iOS mobile builds contain zero third-party ad or tracking SDKs.
01. Account Identity & Authentication
The Pak-LLM Coder IDE and Agent Mission Control offer optional authenticated sessions powered by Sovereign Cloud Authentication (pak-llm.firebaseapp.com). Anonymous users may explore starter templates and downloads freely.
02. Custom Domain Registration & WHOIS Privacy
When purchasing custom apex domains (.com, .pk, .com.pk, .ai, .tech, etc.) through our Sovereign Registrar Engine, registrant contact details (name, email, phone) are securely passed to upstream ICANN registrar rails solely to fulfill domain ownership legal requirements.
- Free WHOIS Privacy Guard: Enabled by default on all eligible TLDs to shield your personal email, address, and telephone numbers from public registries.
- DNS Host Records: Records you configure (@ A, www CNAME, TXT, MX) are published to authoritative nameservers for public internet routing.
- 10% Platform Service Fee: Transparently calculated at checkout without hidden recurring surcharges.
03. Information We Store Per Workspace
Workspaces created in the Pak-LLM Coder IDE store isolated project structures up to 500 MB per container. Data includes:
- Workspace Files: Source files, assets, and project configurations stored locally in browser storage and synced to cloud workspaces upon save.
- Subscription & Quota Metadata: Active plan metadata (Awam Free / Freelancer Pro / Agency), monthly token usage, and replenishment timestamps.
- Custom Subdomains: Deployed endpoints under
.coder.pak-llm.comor linked apex domains.
04. Encrypted SMTP Notifications & Invoices
When a user activates a package, buys a custom domain, or submits feedback, transactional emails (invoices, receipt confirmations, quota warnings, and admin fulfillment alerts) are dispatched via encrypted TLS/SSL SMTP transport. We do not sell or share email addresses with advertisers.
05. AI Prompt Privacy, Pak-LLM CLI & Model Routing
Inference requests from the Web IDE, Desktop Apps, and Pak-LLM CLI are routed directly to our fine-tuned coder model (saadfarrukh191/PakLLM_Coder). In-flight prompts are protected with HTTPS TLS 1.3 encryption. For CLI operations, all file views, diff patching, and bash commands run strictly on your local developer machine. We enforce a zero-data-retention policy—your private code is never logged or used to retrain public base models without explicit authorization.
06. Subprocessors & Infrastructure Transparency
| Subprocessor | Purpose | Data Handled | Region |
|---|---|---|---|
| Global Edge Network | Edge CDN, WAF, Nonce CSP & Serverless Hosting | HTTP requests, transient IP logs, edge cache | Global Anycast / Singapore |
| Swich Fintech Pvt. Ltd. | Domestic 0% Riba Payment Gateway | Order Ref, Amount (PKR), Channel (JazzCash/EasyPaisa) | Karachi, Pakistan |
| ICANN Registrar Rails | Custom Apex Domain Registration & DNS Hosting | Domain SLD/TLD, WHOIS contacts (Protected by WhoisGuard) | Global ICANN Registries |
| Lark Mail (SBF Consultancy (Pvt.) Ltd.) | Encrypted SMTP Transactional Delivery | Invoices, Purchase alerts, Bug reports | Encrypted Cloud SMTP |
07. Explicit Data Retention Schedules by Category
In compliance with ISO/IEC 27001, NIST CSF 2.0, and statutory privacy frameworks, Pak-LLM Coder enforces strict, category-specific data retention schedules:
| Data Category | Scope & Description | Retention Period | Purge & Deletion Mechanism |
|---|---|---|---|
| Account Identity | User ID, email, preferences, authentication state | Duration of active account lifecycle | Purged within 30 days of account deletion request |
| Workspace Files | Mounted files, AST structures, code editor buffers | Active session + 90 days post-dormancy | Instant deletion via IDE or automated secure purge |
| Subscription / Quota Metadata | Token usage counters, tier limits, cycle reset dates | Active billing period + 12 months | Aggregated anonymized rollover |
| Payment Records | Swich transaction IDs, invoice refs, tax VAT logs | 5 years (Statutory SBP & tax compliance rules) | Encrypted compliance vault archive |
| Transient Logs & Security Telemetry | CSP violation reports, IP hashes, firewall triggers | 30 days rolling window | Automatic edge log rotation & sanitization |
08. Programmatic SDK Telemetry, Client-Side Secret Redaction & Ethical Auditing
When utilizing the sovereign developer SDKs (@pak-llm/sdk for TypeScript / Node.js and pakllm-coder for Python), developers benefit from automated client-side protection:
- Client-Side Secret Redaction: High-entropy API keys, database credentials, JWTs, and private tokens are automatically scrubbed on your local client before in-flight requests leave your environment.
- Zero Source Code Retention: Code synthesized or inspected via the SDK executes in your local process or browser sandbox; source code is never permanently stored on Pak-LLM servers.
- Behavioral Usage Telemetry: Aggregated feature interactions (e.g. playground runs, code copy events, SDK package downloads) are transmitted asynchronously to the Pak-LLM monitoring hub (
pak-llm.com/api/admin/coder/webhook) to verify quota limits and platform health. - Ethical & Islamic Law Sentinel: In accordance with Islamic Jurisprudence and the Pakistan Electronic Crimes Act (PECA 2016), automated perimeter heuristics prevent misuse for unauthorized cyber exploitation, usury (Riba), gambling/betting software (Maysir), obscenity (Fahsha), fraudulent Ponzi schemes (Gharar), or blasphemy. Infringement attempts are immediately blocked and logged to the admin dashboard with project and subsystem segregation.
10. In-App User Prompt Telemetry, Project Overview Gathering & Central Admin Segregation
To verify software quality, analyze developer workflow behavior, monitor agent performance, and uphold strict Islamic Shariah and ethical standards, Pak-LLM Coder collects and aggregates prompt and project metadata across all four application surfaces (Web IDE, Enterprise Mission Control, Decoupled CLI, and Sovereign SDK):
- User Prompt Collection: Prompts submitted by users to synthesize code, scaffold microservices, or orchestrate autonomous missions are captured along with timestamp, app surface identifier, and developer session ID. Sensitive credentials (API keys, JWTs, private keys) are scrubbed via client-side secret redaction prior to ingestion.
- Project Overview Telemetry: High-level project metadata—including workspace name, technology stack (e.g. React 19, Python FastAPI, Three.js, Rust Axum, Go Fiber), total file count, and file summary lists—is collected to provide comprehensive visibility into active developer workloads.
- Segregated Admin Webhook Relay: All prompt and project overview telemetry is transmitted asynchronously via authenticated webhooks and APIs (
/api/admin/coder/webhook) to the centralized Pak-LLM Admin Dashboard, segregated strictly under thepakllm-coderproject namespace. - Immediate Rejection of Illegal & Immoral Software Prompts: Prompts attempting to generate illegal, deceptive, usurious, or immoral applications (such as gambling/casinos, compound usury calculators, adult pornography, fake bank transfers, phishing, darknet drug vendors, or cyberattacks) are immediately rejected at the perimeter with zero model ingestion.
- Administrative SMTP & Webhook Alerts: Every detected violation triggers an automated real-time incident notification dispatched simultaneously to the Pak-LLM Admin Dashboard and via secure SMTP email alerting to the security administration (
[email protected]).
09. Native Mobile Applications (Android & iOS) Device Privacy & Permissions
The downloadable Pak-LLM Coder mobile and tablet apps (.apk / .aab for Android and .ipa for iOS) are engineered according to strict privacy-by-design standards:
- Zero Advertising Trackers: The mobile apps never request, read, or transmit device advertising identifiers (Google Advertising ID / AAID or Apple Identifier for Advertisers / IDFA). Zero third-party analytics or ad SDKs are embedded.
- Scoped Filesystem Access: The apps use Android Storage Access Framework (SAF) and iOS Document Picker with security-scoped bookmarks. The app only accesses files and project folders explicitly selected by the user.
- Hardware Biometrics Isolation: When biometric authentication is enabled (Face ID, Touch ID, or Android BiometricPrompt), encryption keys are isolated exclusively inside the hardware security module (Apple Secure Enclave or Android Keystore StrongBox). Biometric measurements never leave device silicon.
- Local WASM Execution: Interactive code evaluations (Python/Node/WASM) execute on-device without remote round-trips unless cloud AI inference is explicitly requested by the user.
10. Data Rights & Data-Subject Access / Deletion Request (DSAR) SLA
You possess full statutory rights under applicable privacy frameworks to access, rectify, export (data portability), or permanently delete your personal information and code workspaces.
All verifiable data-subject access, deletion, and export requests submitted to our official privacy desk are acknowledged within 24 business hours and fulfilled within a strict service level agreement (SLA) of 30 calendar days with zero administrative processing fees.
Parent Architecture: SBF Consultancy (Pvt.) Ltd. (sbf-consultancy.net)
11. Third-Party Subprocessors & Hybrid Data Residency
We maintain full transparency regarding where your data flows. Client-side WASM execution and local CLI operations run completely on your physical machine with zero code storage. Cloud metadata, identity tokens, and domestic settlements use vetted, minimal subprocessors:
- SBF Consultancy (Pvt.) Ltd. Sovereign Enclave (Pakistan 🇵🇰): Primary inference cluster running fine-tuned PakLLM_Coder 32B weights with zero code storage.
- Swich Fintech Pvt Ltd (Pakistan 🇵🇰): SBP PSO/PSP regulated domestic settlement in PKR via JazzCash, EasyPaisa, 1Link.
- Global Edge Network (USA / Global Edge): Frontend hosting, anycast CDN routing, and edge DDoS defense.
- Google Cloud / Firebase (USA / EU): User identity authentication, password hashing, and App Check protection.
- Neon Inc. (USA): Serverless PostgreSQL storage for user account tiers and encrypted workspace metadata.
- Hugging Face Inc. (USA / EU): Secondary fallback model checkpoint repository.
For complete data categories, jurisdictions, and retention periods, see our detailed Subprocessors Directory in the Compliance Charter.
12. Mission Control Artifacts, Workers & Audit Data
The Enterprise Mission Control console is client-first. Architecture indexing, context slicing, AST validation and virtual worker snapshots execute in your browser session; repository contents are not uploaded unless you explicitly invoke a server verification or artifact endpoint.
- Mission artifacts: Plan & Spec, Atomic Patch Set and Verification Matrix bundles are content-addressed and immutable. They remain in the active browser session unless you export JSON or explicitly persist them to the server artifact store.
- Worker isolation: documentation, test-authoring, security-review and refactor workers operate on isolated virtual branch snapshots. Base workspace files are not modified until a verified, human-approved merge.
- Audit ledger: mission transitions are recorded in a hash-chained trail. Secret-like keys and values are redacted recursively before storage; the admin audit endpoint requires the sovereign admin key.
- Server verification: when you run server-side verification, files are path-normalized, size-limited (300 KB per file, 2 MB total, 250 files), rate-limited (30 requests/minute) and never retained after the response.
- Real Git worktrees: disabled by default. If an administrator enables them, operations are confined to a configured trusted root and executed with an argv-only Git call; branch names and worktree paths are validated before use.
- Data-subject rights: artifact exports and audit data can be deleted through the standard DSAR process described in Section 06, within the same 30-day SLA.
Privacy & Data FAQs
Frequently asked questions regarding code privacy, domain WHOIS data, and zero-retention policies.
No by default. Architecture indexing, BPE-budgeted context slicing, AST-guarded patch validation and virtual worker snapshots run in your browser session. Repository files leave the device only when you explicitly run a server-side verification or persist an artifact bundle; those requests are path-validated, size-limited, rate-limited and not retained after the response.