Skip to content

Architecture

Runtime model

Scibudy is organized as a Python-first runtime with a lightweight npm bootstrap wrapper.

Layers

  • research_mcp/
  • MCP server
  • CLI
  • library/catalog management
  • analysis engine
  • local model worker clients
  • web/
  • React/Vite UI source
  • built static assets
  • bin/scibudy-install.mjs
  • bootstrap wrapper for cross-device install

State model

Source tree and runtime state are separate:

  • source tree: code, docs, workflows, examples
  • app home: .env, state DB, install state, UI assets, analysis outputs

Analysis model

Retrieval pipeline:

  1. lexical ranking
  2. dense embedding recall
  3. reranking
  4. summary/report generation

Local models

Dedicated local model workers are isolated from the main runtime so that heavy GPU dependencies do not destabilize the main CLI/MCP environment.