Chitragupta
A memory-oriented AI agent platform that treats cognition as an engineering discipline — most of it running at zero LLM cost.
- 17 packages, 11,500+ tests, zero failures
- Bi-temporal knowledge graph with time-travel queries
- Sleep consolidation — five-phase dream cycles at 2am
- 38 MCP tools; retrieval fuses BM25, vectors, and GraphRAG
Your AI assistant forgets everything. Chitragupta fixes that.
It is the center of everything else I build: a single-writer daemon that owns memory, identity, attention, and intention for every agent surface — CLI, HTTP, MCP, dashboard, menubar. The cognitive layer (Chetana) runs on heuristics, not API calls, so the platform thinks continuously without burning a token.
Every edge in its knowledge graph carries an epistemological type — six Pramana categories, the Vedic theory of valid knowledge, applied to retrieval ranking. The name is not decoration; the tradition spent millennia modeling the mind, and the architecture borrows from it deliberately.
Private for now — closer to the workshop than the storefront. Ask me about it.