Architectural Security
Security & Repository Privacy
How LLMGRAPH protects your codebase while delivering persistent semantic context to coding agents.
1. Read-Only Git Repository Access
LLMGRAPH requests read-only repository permissions. We do not require write access or pull request manipulation permissions.
2. Temporary Acquisition & Immediate Code Destruction
Repository source code is cloned into an isolated ephemeral worker for AST parsing. Once the semantic graph is generated, the raw source code checkout is destroyed.
3. Persistent Derived Semantic Graph Only
LLMGRAPH stores derived structural topology (symbol identifiers, function declarations, line ranges, and call edges). We do not store raw implementation source code in our semantic database.
4. Credential Isolation & Model Invisibility
Git credentials and auth tokens are not model-visible MCP parameters and are never passed into agent prompts or context payloads.