Memra

AI agent memory infrastructure for ai agents

Your AI tools forget everything. Fix that.

Persistent memory for Claude Code, Cursor, and every AI tool in your stack. Local-first. Cloud-synced. EU-hosted.

ai agent memory • semantic recall • vector embeddings • graph memory • pii masking • privacy-first • eu-hosted • <200ms p50 target • 99.5% uptime target • mcp native • rest api • ai agent memory • semantic recall • vector embeddings • graph memory • pii masking • privacy-first • eu-hosted • <200ms p50 target • 99.5% uptime target • mcp native • rest api • ai agent memory • semantic recall • vector embeddings • graph memory • pii masking • privacy-first • eu-hosted • <200ms p50 target • 99.5% uptime target • mcp native • rest api • ai agent memory • semantic recall • vector embeddings • graph memory • pii masking • privacy-first • eu-hosted • <200ms p50 target • 99.5% uptime target • mcp native • rest api •

‹ 01 ›

> what is memra

What is memra AI agent memory?

Memra is memory infrastructure for AI agents. Drop in a REST API or MCP server and your agents gain persistent, long-term memory, semantic recall, graph context, and PII masking. EU-hosted and privacy-first — it’s the AI memory agent layer built for teams who can’t afford to repeat themselves.

✓ ai agent memory ✓ persistent memory api ✓ memory layer ✓ llm memory ✓ mcp-native ✓ eu-hosted ✓ claude code ✓ cursor

‹ 02 ›

> integration

Two lines to remember.One line to recall.

works with any ai tool that supports mcp. or call the rest api directly.

# store a memory
curl -X POST https://usememra.com/api/v1/memories \
  -H "Authorization: Bearer memra_live_..." \
  -H "Content-Type: application/json" \
  -d '{
    "content": "User prefers dark mode and concise responses",
    "type": "preference"
  }'

# recall relevant memories
curl -X POST https://usememra.com/api/v1/memories/recall \
  -H "Authorization: Bearer memra_live_..." \
  -H "Content-Type: application/json" \
  -d '{"query": "user preferences"}'_

also available: python sdk • php sdk • mcp server • rest api

‹ 03 ›

> metrics

Built for production

real infrastructure. real performance. real privacy.

EU

hosted

hetzner helsinki, finland.
sovereign mode: zero non-eu subprocessors.

<200ms

p50 latency

semantic search
in milliseconds.

5

factor recall

vector + temporal +
importance + recency + graph.

99.5%

uptime target

single-tenant
eu infrastructure.

11

endpoints

complete rest api.
store, recall, search, export.

‹ 04 ›

> data flow

How memory works

01

store

Your AI agent sends memories via REST API or MCP. Content is written atomically to the flat-file engine, metadata indexed in PostgreSQL, and embeddings queued asynchronously.

POST /v1/memories → queue → embed → index

02

process

Optional PII masking via Presidio + spaCy detects sensitive data in 24 EU languages. Embeddings are generated via OpenAI and cached with tenant-aware Redis keys for privacy.

pii scan → embed → cache → graph extract

03

recall

5-factor intelligent recall scores every memory against your query using vector similarity, temporal relevance, importance weighting, recency decay, and entity graph context. Built for sub-200ms p50 recall targets.

query → 5-factor score → ranked results → <200ms

‹ 05 ›

> pricing

Simple, transparent pricing

see full comparison →

hobby

Free

forever

• + 2,500 memories
• + 5,000 recalls/mo
• + 1 project
• + 2 API keys
• + 30 req/min

dev

EUR 39/mo

per account

• + 25,000 memories
• + 50,000 recalls/mo
• + 5 projects
• + intelligent recall
• + graph memory

team

EUR 399/mo

per account

• + unlimited memories
• + unlimited recalls
• + 100 projects
• + intelligent recall
• + graph memory