Security
Esta página aún no está disponible en tu idioma.
Privacy by design
Section titled “Privacy by design”polyglot-gpu runs entirely on your local machine. No data is sent to any external service.
Threat model
Section titled “Threat model”Data touched
Section titled “Data touched”- Text sent to the local Ollama API at
localhost:11434 .polyglot-cache.jsonsegment cache in the working directory
Data NOT touched
Section titled “Data NOT touched”- No files outside the working directory
- No browser data, OS credentials, or system files
- No environment variables beyond
POLYGLOT_MODELandPOLYGLOT_CONCURRENCY
Network
Section titled “Network”- HTTP to
localhost:11434only - Zero external or internet egress
- No DNS lookups, no telemetry endpoints
No telemetry
Section titled “No telemetry”- No data collection or transmission of any kind
- No usage analytics, crash reports, or phone-home behavior
No secrets
Section titled “No secrets”- No API keys, tokens, or credentials required or stored
- No authentication mechanism — Ollama runs locally without auth
Cache security
Section titled “Cache security”The segment cache (.polyglot-cache.json) stores translated text segments locally. Path traversal protection uses Python’s Path.relative_to() to prevent cache files from escaping the intended directory.
If your source text contains sensitive content, be aware that translations are cached in plaintext. Delete the cache file to remove stored translations:
rm .polyglot-cache.jsonDependency security
Section titled “Dependency security”polyglot-gpu depends on:
httpx— HTTP client for Ollama communicationmcp[cli]— FastMCP server frameworkpython-Levenshtein— Fuzzy string matching for cache
CI runs pip-audit on every push to detect known vulnerabilities in dependencies.
Reporting vulnerabilities
Section titled “Reporting vulnerabilities”Report security issues to: 64996768+mcp-tool-shop@users.noreply.github.com