Your Code, Your Control

Forge reads your code to build a map of its structure. Here's exactly what that means for your data.

What happens when you scan a repository

  1. Parse — Forge reads your source files using language-native parsers (libclang for C++, tree-sitter for Python and TypeScript)
  2. Map — Functions, types, files, and their relationships become nodes and edges in a structural map of your code
  3. Isolate — Your map is stored in your own tenant space, separate from every other user
  4. Query — You ask questions and get structural facts via REST API, Python CLI, or MCP

What we store

  • Function names, signatures, and file paths
  • Call graphs and dependency chains
  • Type relationships and module structure
  • Structural patterns discovered in your code
  • Thermal state (query frequency and recency)

What we don't

  • Your LLM API keys are either per-request (memory only) or optionally stored encrypted (AES-256-GCM) — your choice
  • No model training on your code — the engine is deterministic
  • No sharing between tenants — each map is fully isolated

The map captures how your code is organised — functions, their relationships, and structural patterns. Per-request LLM keys exist only in memory. Stored keys are encrypted at rest with AES-256-GCM and can be deleted from your dashboard at any time.