Chitragupta gives AI agents persistent memory. Your coding assistant forgets everything between sessions — Chitragupta fixes that.

It works as an MCP server, a CLI, or a library you import directly.

Install

npm install @yugenlab/chitragupta

Use as MCP Server

Give any AI agent (Claude Code, Codex, etc.) persistent memory across sessions.

# One-command setup in your project
npx chitragupta init

This creates .mcp.json and teaches the agent when to use Chitragupta's 28 tools — memory search, recall, fact extraction, handover, and more.

Manual MCP setup

Add to your .mcp.json:

{
  "mcpServers": {
    "chitragupta": {
      "command": "npx",
      "args": ["chitragupta-mcp"],
      "env": {
        "CHITRAGUPTA_MCP_PROJECT": "/path/to/your/project"
      }
    }
  }
}

Use as CLI

# Interactive mode
npx chitragupta

# Search memory
npx chitragupta recall "what did we decide about auth?"

# Run coding agent
npx chitragupta code "fix the login bug"

Use as Library

import { createSession, addTurn, recall, DatabaseManager } from "@yugenlab/chitragupta"

// Initialize
const db = new DatabaseManager();

// Create a session
const session = await createSession(db, {
  project: "/path/to/project",
  provider: "anthropic",
});

// Search across all memory
const answer = await recall(db, "how did we implement caching?");

Sub-path Imports

Each subsystem is available as a separate import:

import { ... } from "@yugenlab/chitragupta/core"    // Types, errors, config
import { ... } from "@yugenlab/chitragupta/smriti"   // Memory, sessions, GraphRAG
import { ... } from "@yugenlab/chitragupta/swara"    // LLM providers, routing
import { ... } from "@yugenlab/chitragupta/anina"    // Agent runtime, consciousness
import { ... } from "@yugenlab/chitragupta/tantra"   // MCP server/client
import { ... } from "@yugenlab/chitragupta/dharma"   // Policy, security rules
import { ... } from "@yugenlab/chitragupta/yantra"   // Tool system
import { ... } from "@yugenlab/chitragupta/vayu"     // Workflow DAG engine
import { ... } from "@yugenlab/chitragupta/sutra"    // IPC, actor mesh
import { ... } from "@yugenlab/chitragupta/niyanta"  // Orchestrator
import { ... } from "@yugenlab/chitragupta/netra"    // Vision
import { ... } from "@yugenlab/chitragupta/ui"       // Terminal UI
import { ... } from "@yugenlab/chitragupta/darpana"  // LLM API proxy
import { ... } from "@yugenlab/chitragupta/vidhya-skills" // Skill discovery

What It Does

• Persistent memory — sessions, turns, and facts survive across conversations
• Unified recall — single query searches FTS5, GraphRAG, day files, and memory
• Real-time fact extraction — detects personal facts from conversations at zero LLM cost
• Day consolidation — daily summaries across all projects and providers
• GraphRAG — knowledge graph with bi-temporal edges and personalized PageRank
• Hybrid search — BM25 + vector + graph fusion with learned weights
• Sleep consolidation — 5-phase dream cycle makes the agent smarter between sessions
• Behavioral crystallization — stable habits detected via Bayesian change-point detection
• Multi-agent deliberation — structured councils with fallacy detection
• 28 MCP tools — memory, file ops, shell, search, and self-awareness

Requirements

• Node.js >= 22
• At least one AI provider (Anthropic, OpenAI, Google API key, or Ollama for local)

Links

• GitHub
• Getting Started
• Architecture
• Changelog

License

MIT © 2026 Srinivas Pendela