Skip to content

Architecture Overview ​

This page describes FlowMind at a product level only: capability layers and data flow. It does not include source paths, module names, IPC contracts, or database design—those details are confidential and are not published in public documentation.

Product scope ​

FlowMind is a desktop application security workbench that integrates in a single process:

  • Traffic proxy & capture: HTTP/HTTPS/WebSocket
  • Analysis & actions: forward, intercept, replay, fuzz
  • Security testing: built-in rules + extensible scan plugins
  • AI assistance: chat, tools, knowledge base, project context
  • Deliverables: project isolation, findings, report export

Core workflows run locally without requiring cloud services.

Logical layers (conceptual) ​

mermaid
flowchart TD
    UI[User interface] --> Core[Application core]
    Core --> Proxy[Proxy & protocols]
    Core --> Store[Local storage]
    Core --> Scan[Scanner & plugin runtime]
    Core --> AI[AI & orchestration]
    Proxy --> Store
    Scan --> Store
    AI --> Store
    Core --> UI
LayerRole
UIPresents traffic, findings, settings, AI; no duplicated business rules
CoreOrchestrates subsystems, entitlements, project context
ProxyListen, MITM, connection and message handling
StorageFlows, findings, config, sessions
Scanner & pluginsBuilt-in rules + Rhai / declarative extensions
AIMulti-provider chat, tools/MCP, memory and RAG

Data & privacy ​

  • Flows, findings, AI sessions, etc. are stored locally by default.
  • Certificates, config, and plugin workspaces live under the app data directory (Settings → General).
  • Data sent to external LLMs or MCP depends on AI settings; use local models or enterprise gateways in sensitive environments.

Extension model ​

Public implementation-level documentation is limited to scan plugins:

Plugins run in a controlled runtime and interact through documented inputs/outputs without requiring host source access.

See also ​

Maintainers & authorized contributors

Internal architecture, IPC, and data model docs are available through authorized channels only. Please do not request implementation details in public issues or docs.