NewNew: 60dB Judge answers 214 of 231 JevBench decisions — 15 more than Jev, at $0.010 per 1M input tokens.See the benchmark
60db Logo

AI Memory ThatNever Forgets

60dB SmritiPersistent Memory for Voice AI

60dB Smriti is a persistent memory API for voice AI. It scored 92.4% on LongMemEval-s with about 185 ms median recall. Ingest documents, search with hybrid vector + full-text retrieval, and assemble LLM-ready context through a simple API.

Get Started FreeRead the Docs
Vector SearchDocument ExtractionCollectionsContext AssemblyAPI

Everything you need for persistent AI memory

From ingestion to retrieval, a complete memory layer for your AI applications.

Hybrid Search

Vector + full-text search with configurable alpha weighting. Get the precision of keyword search with the understanding of semantic search.

91+ Document Formats

Extract text from PDFs, DOCX, images with OCR, spreadsheets, and more. Upload any document and we handle the rest.

Collections

Organize memories into personal, team, knowledge, and hive collections. Fine-grained access control keeps data secure.

Context Assembly

Get LLM-ready context with sources, graph context, and timeline. Drop it straight into any prompt with full attribution.

Batch Ingestion

Ingest up to 100 memories at once. Automatic retry queuing ensures nothing gets lost, even during upstream failures.

Smart Chunking

Configurable chunk size and overlap for document extraction. Up to 100 chunks per document with intelligent splitting.

How it works

Three steps to persistent AI memory.

Step 01

Ingest

Feed documents, text, or batch data into your memory store. Supports 91+ formats with automatic text extraction and smart chunking.

Step 02

Search

Query with hybrid vector + full-text retrieval. Configurable ranking, reranking, and recency bias across all your collections.

Step 03

Retrieve

Get LLM-ready context with sources, graph context, and timeline. Drop it straight into any prompt for grounded AI responses.

Built for real-world AI

From chatbots to research tools, memory makes your AI applications smarter.

Customer Support Bots

Store conversation history and knowledge base articles. Your bot recalls past interactions and resolves issues with full context.

Research Assistants

Ingest papers, extract insights, and retrieve relevant context across your entire research corpus with hybrid search.

Personal AI Agents

Give agents persistent memory across sessions. They remember preferences, past decisions, and evolving context.

Knowledge Management

Build a searchable team knowledge base from documents, notes, and conversations. Organized with collections and access control.

Frequently asked questions

You can ingest plain text, structured data, and documents in 91+ formats including PDF, DOCX, XLSX, images (with OCR), and more. Each memory can be up to 100,000 characters.

Hybrid search combines vector (semantic) search with full-text (keyword) search. You can control the balance with the alpha parameter — higher alpha favors semantic results, lower alpha favors exact keyword matches.

Collections let you organize memories into logical groups. You can create personal collections (private to you), team collections (shared with your workspace), knowledge bases, and hive collections (shared across your entire tenant).

Memory uses your plan credits, and pay-as-you-go rates once they run out. Ingestion is billed per 1K characters ($0.0001), document extraction per MB ($0.003), search queries at $0.30 per 1,000 queries, and context assembly at $0.50 per 1,000 queries. You only pay for what you use.

Context assembly takes your search results and formats them into LLM-ready context. It includes source attribution, graph context, and timeline information — ready to drop into any prompt for grounded AI responses.

Yes. The Free plan includes 3,000 credits every month that work across all 60db services including memory. No credit card required.