🚧   This page is coming soon
✓   Copied to clipboard
NEWv0.4.2 is live — The latest performance and feature updates are now available.
Menu
Libraries
⚡ @manasdb/core🔌 @manasdb/mcp-server
Memory Layer for AI Applications

AI memory thatlives on your database.

ManasDB Core is the Node.js-native SDK for AI memory — local embeddings, hybrid retrieval, polyglot storage, and MCP-native integration. No cloud lock-in. No API key to start.

●v0.4.2 Core stableApache 2.0 + Commons ClauseNode.js ≥ 18MongoDB · PostgreSQL
29×
Faster repeated queries with Redis cache
50%
API cost reduction via deduplication
+9.7%
Better recall accuracy vs raw stack
2
Production-stable npm packages

Permanent memory in 10 seconds.

Plug your own database, choose your embedding model, and start memorizing everything. No complex setup, no cloud lock-in.

index.js
1
2
3
4
5
6
7
8
9
10
11
12
import { ManasDB } from '@manasdb/core';
const memory = new ManasDB({
uri: process.env.MONGODB_URI,
modelConfig: { source: 'transformers' },
});
await memory.init();
await memory.absorb('ManasDB Core is the Node.js memory layer for AI.');
const results = await memory.recall('What is ManasDB Core?');
console.log(results[0].metadata.matchedChunk);
output● live
result[0]
contentId"cnt_7f3a9b2c…"
score0.9381
metadata
"Primary mirror: 6.5m,
18 beryllium segments."
_trace
cacheHitfalse
rrfMerged14
finalScore0.9381
tokens12
costUSD$0.00024
Libraries

Two packages.
Everything you need.

Start with @manasdb/core and bring your own database. Add the MCP server to give any AI assistant persistent memory — no custom integration needed.

⚡
@manasdb/core
Core SDK
STABLE

The foundation. absorb(), recall(), reasoningRecall(). Polyglot broadcasting to MongoDB and PostgreSQL, hybrid RRF+MMR retrieval, two-tier Redis caching, PII Shield, and full cost telemetry.

absorb()recall()reasoningRecall()PII ShieldTelemetry
Docs →npm ↗
v0.4.2
🔌
@manasdb/mcp-server
MCP Server
STABLE

Give Claude Desktop and Cursor permanent memory in 60 seconds. An interactive setup wizard generates your config. Exposes memorize, recall, and forget as native MCP tools.

memorizerecallforgetClaude DesktopCursor
Docs →npm ↗
v0.1.1
ManasDB Core vs Traditional Stack

Built for Node.js.
Built for privacy.

A traditional RAG stack requires complex glue code and vendor-locked cloud APIs. ManasDB Core is Node.js-native — your memory layer stays on your infrastructure.

Integration
Development runtime
Traditional Stack
Manual Glue Code
ManasDB Core
Node.js native ✓
Data Privacy
Where your data goes
Traditional Stack
Vendor Lock-in
ManasDB Core
Stays on your server ✓
Local Embeddings
No API key needed
Traditional Stack
Requires Cloud APIs
ManasDB Core
Ollama / Transformers ✓
Hybrid Search
Dense + sparse + RRF
Traditional Stack
DIY Implementation
ManasDB Core
RRF + MMR built-in ✓
Redis Caching
29x faster repeated queries
Traditional Stack
Complex Setup
ManasDB Core
Tier 1 + Tier 2 ✓
Tree Reasoning
Hierarchical recall
Traditional Stack
Manual Logic
ManasDB Core
reasoningRecall() ✓
PII Protection
Before data leaves server
Traditional Stack
Pre-processor DIY
ManasDB Core
Built-in per-field ✓
MCP Integration
Claude · Cursor
Traditional Stack
Partial
ManasDB Core
Working today ✓
Trace Debugging
Full pipeline audit
Traditional Stack
Log parsing
ManasDB Core
Every recall() ✓
On LangChain & LlamaIndex: ManasDB Core operates at the storage layer. These frameworks are excellent for chaining LLM calls — ManasDB Core is the memory backend that plugs into them. Complementary, not competing.
Capabilities

Everything ManasDB Core
does for your memory layer.

Everything you'd bolt on manually — reranking, caching, deduplication, PII filtering, cost tracking — built directly into the SDK.

⚡
Polyglot Broadcasting
Write once, sync to MongoDB and PostgreSQL simultaneously. Migration, cross-region replication, or disaster recovery.
absorb() → all providers
🔀
Hybrid Retrieval
Dense ANN vector search fused with sparse keyword search via Reciprocal Rank Fusion, diversified with Maximal Marginal Relevance.
RRF + MMR
🌲
Tree Reasoning
Maps chunks into Document → Section → Leaf hierarchy. Returns the highest-scoring section's leaves for deep structured answers.
reasoningRecall()
⚡
Two-Tier Cache
Tier 1 shared Redis across servers. Tier 2 in-memory LRU. Both short-circuit the DB when cosine similarity ≥ 0.95.
29x faster
🛡️
PII Shield
Regex-based redaction of emails, phone numbers, SSNs, and custom patterns before any text hits your database.
Before storage
🔍
Trace Debugging
Every recall() emits a _trace: cache hit status, PII tokens scrubbed, candidate counts, fallback triggers, final score.
_trace on every call
📊
Cost Telemetry
Tracks tokens, API cost, and latency savings in _manas_telemetry on your own database. Never leaves your server.
npx manas stats
📊
Sentinel Micro-Index
Dual-layer storage: chunk-level vectors for broad recall, sentence-level for QA. Boosts short-form precision ~30%.
mode: 'qa'
🔌
Custom Drivers
Plug any air-gapped or corporate embedding model. ManasDB becomes your standard interface over the entire embedding stack.
source: 'custom'
Performance

ManasDB Core Benchmarks

Run npx manas benchmark against your own URIs. These are real numbers from a free Atlas M0 cluster.

29×
Faster repeated queries
Redis Tier 1: complex QA 120ms → 4ms
-97%
Latency reduction
310ms raw stack → 9ms (MongoDB)
-50%
API cost reduction
SHA256 dedup + float16 compression
Metric
Raw
ManasDB
Delta
Latency avg310ms9ms-97%
Absorb time1200ms673ms-44%
API cost / 10k$0.024$0.012-50%
Recall accuracy82.4%92.1%+9.7%
Dedup / CacheNoneSHA256✓
Get started today

Own your memory.
Start with ManasDB Core today.

Free MongoDB Atlas cluster + local embeddings. Your data never leaves your server.