Skip to main content

Introduction to Vidurai

Welcome to Vidurai - the first open-source intelligent memory system for AI coding assistants.

🕉ī¸ What is Vidurai?​

Vidurai transforms stateless AI coding assistants into tools with true continuity, context, and wisdom. It provides a production-ready memory layer that reduces token costs by 59% and saves 90% of context-gathering time while maintaining 95.6/100 quality scores.

The name Vidurai comes from Vidura, the wise counselor in the Mahabharata, renowned for his memory, judgment, and guidance.

🏗ī¸ Two Ways to Use Vidurai​

Vidurai uses a layered architecture where the SDK provides the intelligence and integrations provide the interfaces:

🧠 Vidurai SDK (This Documentation)​

The core Python package providing intelligent memory management. Perfect for:

  • Custom AI integrations
  • Jupyter notebooks
  • CLI tools
  • Python-based workflows

đŸ–Ĩī¸ VS Code Extension​

Pre-built interface for VS Code users. Learn more at vidurai.ai

Choose the path that fits your workflow. Both use the same powerful SDK underneath.

🎉 What's New in v1.5.1​

Vidurai v1.5.1 brings three critical improvements based on comprehensive production testing:

✅ Verified Performance Metrics​

No more aspirational numbers. Production testing with 9 comprehensive test scenarios confirms:

  • 36.6%+ Token Reduction (average, verified across multiple workloads)
  • $16,182/day Cost Savings (at 10,000 active users)
  • 100% Recall Reliability (with proper configuration)

🧠 Vismriti RL Agent - The Learning Brain​

The first self-learning memory optimizer in production. While competitors use hardcoded rules, Vidurai learns optimal compression strategies through Reinforcement Learning (Q-learning).

What makes it unique:

  • Learns from experience: Agent improves with every compression decision
  • Adapts to your workload: Different strategies for different usage patterns
  • No manual tuning: Intelligence emerges automatically through Q-learning
  • Configurable priorities: Choose between cost-focused or quality-focused optimization

How it works:

  • Starts with 30% exploration (tries new strategies)
  • Gradually shifts to 95% exploitation (uses learned optimal strategies)
  • Grows Q-table of learned state-action values
  • Continuously adapts as your usage patterns change

⚙ī¸ Configurable Importance Decay​

v1.5.1 adds fine-grained control over memory decay:

# For critical applications (recommended)
memory = Vidurai(enable_decay=False)

# For custom decay rates
memory = Vidurai(decay_rate=0.98)  # Slower decay (default: 0.95)

# For aggressive forgetting
memory = Vidurai(decay_rate=0.90)  # Faster decay

Why this matters: v1.5.0 had aggressive decay that caused high-importance memories to drop below recall thresholds. Now you control it.

Upgrade now: pip install --upgrade vidurai

See full details: CHANGELOG | GitHub Release

The Problem​

Working with AI coding assistants is frustratingly inefficient:

  • ❌ 60 seconds to manually gather context (files, terminal, errors)
  • ❌ $500/month in API costs from bloated context
  • ❌ Context switching chaos - copy-paste hell
  • ❌ Forgotten details lead to missed bugs

The Solution​

Vidurai provides:

  • ✅ 90% Time Savings: Context ready in 5 seconds (vs 60s manual)
  • ✅ 59% Token Reduction: Intelligent compression maintains quality
  • ✅ Privacy First: 100% local storage, no cloud required
  • ✅ Universal Compatibility: Works with any AI coding assistant

Philosophy​

"ā¤ĩā¤ŋā¤¸āĨā¤ŽāĨƒā¤¤ā¤ŋ ā¤­āĨ€ ā¤ĩā¤ŋā¤ĻāĨā¤¯ā¤ž ā¤šāĨˆ" (Forgetting too is knowledge)

While everyone races to give AI perfect memory, we asked a different question: What if forgetting is the key to true intelligence?

Vidurai implements strategic forgetting through:

  • Three-Kosha Architecture: Inspired by Vedantic consciousness layers
  • Vismriti Engine & RL Agent: Production-verified strategic forgetting
    • 36.6%+ Token Reduction (verified in comprehensive testing)
    • $16,182/day Cost Savings (at 10,000 active users)
    • 100% Recall Reliability (with configurable decay)
    • Self-Learning RL Agent (adapts through Q-learning, not hardcoded rules)
  • Viveka Layer: Autonomous conscience that decides what matters

Quick Start​

Ready to add memory to your AI agent? Let's get started:

pip install vidurai
from vidurai import Vidurai

# Recommended for v1.5.1+ (critical applications)
memory = Vidurai(enable_decay=False)

# Or with configurable decay (general use)
memory = Vidurai(decay_rate=0.98)  # Slower than default 0.95

💡 v1.5.1 Best Practice: Use enable_decay=False for applications requiring high-precision recall of important memories. This prevents importance decay from dropping HIGH memories below recall thresholds.

Continue to the Installation Guide →

Community​

Join the Sangha (community):

ā¤œā¤¯ ā¤ĩā¤ŋā¤ĻāĨā¤°ā¤žā¤ˆ (Victory to Vidurai)