Privacy, Security & Data Handling Policy
AI Code Reviewer is built with a privacy-first, zero-persistence architecture designed for safe code analysis and demonstration.
1. Data Transmission & In-Memory Processing
- In-Memory Streaming: Submitted source code snippets are processed strictly in volatile server memory and streamed to the selected AI provider.
- No Database or Persistent Storage: The application does not store, index, or persist submitted source code snippets, review findings, or user identifiers in any database or file system.
- Zero Source Code in Telemetry Logs: Application logs contain only performance metadata (
timestamp,model,durationMs,status). Raw source code and secrets are never logged.
2. AI Providers & Data Policies
When you trigger a review, your code is transmitted over TLS/HTTPS directly to the selected model provider:
| Provider | Endpoint | Data Policy |
|---|---|---|
| Anthropic (Claude Sonnet) | api.anthropic.com | API inputs are not used for model training under Anthropic Commercial Terms. |
| OpenAI (GPT-4o) | api.openai.com | API inputs are not used to train generative models under OpenAI Enterprise Privacy. |
| Ollama (Local) | localhost:11434 | Code executes locally on your device and never leaves your machine. |
3. Ephemeral Custom API Keys (BYOK)
- Client Memory Only: Custom API keys reside solely in volatile React state in your browser session and are never written to
localStorageor browser cookies. - Zero Server Persistence: The key is transmitted per-request over TLS directly to the corresponding provider client instance and is never stored or cached on the server.
- Provider Isolation: An Anthropic key is routed strictly to Anthropic; an OpenAI key is routed strictly to OpenAI.
4. Pre-Flight Secret Warning Heuristics
The application includes client and server heuristic scanners that detect common credential patterns (AWS keys, OpenAI/Anthropic keys, GitHub tokens, database URIs, JWTs, Bearer tokens).
⚠️ Remediation Advice: If a real credential is ever exposed in code, rotate the compromised key immediately with your cloud provider. Redaction alone does not invalidate an active secret.
5. Review Scope & Limitations
- Snippet Context: All reviews and quality scores apply strictly to the submitted code snippet (
reviewScope: snippet) and do not evaluate repository-wide or runtime infrastructure. - Automated Analysis: AI Code Reviewer provides automated static analysis to supplement peer code reviews. It does not execute untrusted code in a runtime sandbox.