Skip to content

rvanegas/mdc

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

88 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

mdc — Markdown Chat

Terminal-native LLM research platform Python · Anthropic API


What it is

mdc is a command-line replacement for the standard chat UI, built around the needs of sustained, document-intensive research with language models. It manages structured conversation transcripts, a searchable personal document library, and argument analysis via the Dianoia backend — all from the terminal, integrated with a text editor.

It is the primary interface through which a 3M-word body of philosophical research was developed, indexed, and analyzed across a 200k-token context window.


Why it exists

Standard chat interfaces are designed for discrete, stateless conversations. Research — philosophical, technical, or otherwise — is cumulative. Ideas develop across sessions, earlier conclusions inform later questions, and the ability to retrieve, inject, and reason about prior work is essential.

mdc was built because no existing tool provided: persistent indexed transcript history, flexible document injection into context, argument extraction and analysis, and the kind of fine-grained control over context composition that sustained research requires.


Architecture

Transcript management: Conversations are stored as structured markdown files with consistent naming and metadata. A library indexer generates AI semantic indexes and term-relation maps across the full transcript corpus, with a second co-occurrence supplementation pass for corpus-specific clustering. All transcripts are searchable and injectable into new conversation contexts.

Context composition: mdc gives precise control over what enters the context window. System prompt, injected documents, conversation history, and cached assets are managed as distinct layers. A prompt caching budget manager allocates Anthropic's four cache slots across system prompt and assets to minimize token costs on repeated operations.

Document library: A personal document store with AI-generated indexes. A server-side file ID cache avoids re-uploading unchanged PDFs across sessions. Documents can be injected selectively into any conversation context.

Argument analysis: A subprocess bridge to the Dianoia backend enables argument extraction and evaluation directly from the mdc interface. The mdc argue extract command formalizes an argument from free-form text; mdc argue evaluate assesses each step for validity and strength.

Safety: Versioned backup before every automated file edit. No conversation history or document is modified without a recoverable prior state.

Multi-provider: Supports multiple LLM provider backends with consistent interface abstraction.


Scale

mdc has been used to:

  • Process and analyze 3M+ words of philosophical research
  • Maintain and search a library of 500+ indexed conversation transcripts
  • Generate a comprehensive multi-domain philosophical assessment from a 200k-token context window composed of documents drawn from the library

What it demonstrates

  • Production-quality CLI design for a complex, stateful application
  • Sophisticated context management for long-horizon LLM research
  • RAG-style document retrieval integrated into a conversational workflow
  • Prompt caching strategy across a multi-session research practice
  • Subprocess orchestration for multi-tool AI pipelines

Related projects

  • Dianoia — argument analysis backend
  • Noesis — React frontend for argument evaluation
  • Roxana — collaborative dialectic platform

About

No description, website, or topics provided.

Resources

Stars

Watchers

Forks

Releases

Packages

Contributors

Languages