Skip to content

Nothing escapes a black hole.

Your AI's
long-term memory

Persistent semantic memory for AI agents. One MCP command to connect. Remember context, recall by meaning, learn from every interaction.

Start for free Documentation

No credit card required · 150+ AI agents connected

scroll
Terminal
$ claude mcp add takizen \
  https://mcp.takizen.xyz/mcp \
  --transport http \
  --header "Authorization: Bearer $KEY"
// ~/.cursor/mcp.json

  "mcpServers": 
    "takizen": 
      "url": "https://mcp.takizen.xyz/mcp",
      "headers": 
        "Authorization": "Bearer $KEY"
      
    
  

POST /mcp
Authorization: Bearer mk_...
Content-Type: application/json


  "method": "retrieve",
  "query": "auth session bug"

Features

Built for intelligent agents.
Designed for developers.

Everything you need to give your AI persistent memory, semantic search, and learning capabilities.

Semantic Memory

Store anything. Retrieve by meaning, not keywords. Hybrid search combining vector embeddings and full-text for perfect recall.

  • 1536-dim embeddings via text-embedding-3-small
  • Hybrid semantic + BM25 full-text (RRF fusion)
  • Auto-dedup at 0.92 cosine similarity

Instant Recall

Natural language queries. Results ranked by semantic similarity in milliseconds.

Strength Decay

Memories weaken with time if unused. Active ones stay strong. Inactive ones auto-archive daily.

Memory Graph

Link memories with typed edges: supports, caused_by, part_of.

MCP Native

One command to connect any agent. Standard MCP over HTTP/SSE.

Claude Cursor Cline +more

Namespace Isolation

Each agent gets its own isolated namespace. Zero data leakage by design.

Pricing

Simple, honest pricing

Free tier included. No credit card required.

Free
$0/month
  • 200 active memories
  • 50 recalls / day
  • Semantic deduplication
  • Memory graph links
  • Daily strength decay
  • Weekly consolidation
  • Priority support
Get started free
Most popular
Pro
$9/month
  • 10,000 active memories
  • 1,000 recalls / day
  • Semantic deduplication
  • Memory graph links
  • Daily strength decay
  • Weekly consolidation (LLM)
  • Priority support
Get Pro →

FAQ

Common questions

What is takizen?

takizen is a persistent semantic memory layer for AI agents. It allows your AI to remember context across sessions, recall information by meaning (not just keywords), and learn from past interactions using the MCP protocol.

How does semantic search work?

When you store a memory, takizen generates a 1536-dimensional vector embedding using text-embedding-3-small. When you recall, it compares the query embedding against stored memories using cosine similarity via pgvector, combined with BM25 full-text search for hybrid results.

What is the MCP protocol?

Model Context Protocol (MCP) is an open standard for connecting AI agents to external tools and data sources. takizen implements MCP natively, which means any compatible agent (Claude Code, Cursor, Windsurf, Cline, etc.) can connect with a single command.

Is my data secure?

Yes. Each namespace is completely isolated at the database level. API keys are SHA-256 hashed (never stored in plaintext). All data is stored in EU-hosted Supabase PostgreSQL with Row Level Security enabled. We are GDPR compliant with full data export and erasure support.

What happens when I hit the free tier limit?

The free tier includes 200 active memories and 50 recalls per day. When you reach the limit, new remember operations will return an error. Your existing memories remain accessible. You can upgrade to Pro for 10,000 memories and 1,000 recalls/day.

What is strength decay?

Memories that aren't recalled gradually lose strength over time (daily decay). This naturally surfaces the most useful knowledge while letting stale information fade. Recalled memories are reinforced, keeping them strong. Memories at zero strength are archived to cold storage.