Open Source Client · Zero-Knowledge Encrypted · MCP Compatible

Total recall for
AI agents

Every AI conversation, code edit, and context file — encrypted with CAPRISE zero-knowledge search and ready in milliseconds.

curl -sL engram.ai/install | sh

12ms
median recall
10-50×
compression
0
plaintext on server
50-80%
token savings
$ curl -sL engram.ai/install | sh
✔ Installed engram CLI v0.1.0
$ engram init
→ Opening browser for authentication...
✔ Authenticated as alan@engram.ai
✔ Encryption key generated (stored in keychain)
$ engram migrate --all
Scanning for sources...
✔ OpenClaw (550 sessions)
✔ ChatGPT export (120 sessions)
Encrypting & uploading (client-side)...
[████████████████████████████░░] 92% | 28 chunks/s | ETA 00:12
✔ 4,872 chunks migrated in 2m 48s
$ engram search "WebSocket keepalive fix"
Found 3 results (12ms):
Feb 22 · "Use protocol-level Ping frames, not text pings..."
Feb 19 · "ws.on('pong') handler with 30s heartbeat interval..."
Feb 15 · "Connection drops after 55s — nginx proxy_read_timeout..."

Works with your existing AI tools

Claude ChatGPT OpenClaw Cursor Claude Code Codex Windsurf

Three types of memory,
one system

Everything your AI needs to remember, compressed and encrypted.

💬

Conversation Memory

Store and search years of AI conversations across Claude, ChatGPT, and OpenClaw. "What did we decide about the auth flow three weeks ago?" — answered in milliseconds.

420 bytes/turn compressed
Learn more →

Context Window Optimizer

Replace static file loading with dynamic retrieval. Your AI loads only what's relevant to each message. Cut 50–80% of your token spend.

9,700 → 800 tokens/message
Learn more →
🔧

Code Memory

Every file edit, error, fix, and commit — searchable across sessions. Your coding AI never rediscovers the same bug fix twice.

1 year of coding = 12MB
Learn more →

Built for developers

🔒

Zero-Knowledge Search

CAPRISE-encrypted vectors. Server searches without seeing plaintext.

12ms Recall

HNSW + entity chains return answers in milliseconds across all memory types.

🔗

MCP Protocol

Drop-in MCP server. Works with any MCP-compatible agent or editor.

📊

Usage Insights

See retrieval volume, token savings, and top recalls to tune your agents.

🗜️

10-50× Compression

Schema-compressed vectors shrink storage costs without losing recall accuracy.

🔑

Client-Side Keys

Keys generated and stored in your OS keychain. We never see them.

⏱️

Retention Controls

Configure TTLs, redaction policies, and per-tenant isolation.

🌐

Open Source Client

Audit the encryption boundary yourself. Trust, but verify.

Zero-Knowledge · Peer-Reviewed

Your AI's memory,
not your vendor's.

CAPRISE distance-preserving encryption lets us search your vectors without ever seeing the plaintext. Content is AES-256 encrypted on your machine before upload. We literally cannot read your conversations.

  • Client-side encryption — plaintext never leaves your machine
  • Open-source client — audit the encryption boundary yourself
  • Search accuracy preserved — CAPRISE maintains distance ordering
  • Based on peer-reviewed research (arXiv:2601.12331)
# What your machine sends:
vector: [0.847, -0.331, 0.129, ...] ← CAPRISE
content: "a8f3...b2c1" ← AES-256
# What our server sees:
vector: searchable ✓ (distances preserved)
content: opaque blob ✗ (can't decrypt)
# What you get back:
→ Decrypted locally on your machine
→ Your conversations, readable

Three steps to total recall

From install to searchable memory in under 5 minutes.

1

Capture & Encrypt

Conversations, code events, and context files are compressed and encrypted client-side. Keys never leave your machine.

curl -sL engram.ai/install | sh
2

Search Encrypted Vectors

The server indexes CAPRISE vectors and answers queries without decrypting your data. HNSW + entity chains.

engram search "deploy steps"
3

Retrieve & Inject

Minimal relevant snippets are returned to your agent, keeping context windows lean and costs down.

→ 3 results (12ms)

How Engram compares

The only AI memory with zero-knowledge encrypted search.

Engram Mem0 Zep Supermemory
Zero-knowledge encrypted search
Client-side encryption
10-50× compression
Token savings (50-80%)
Code memory
Multi-AI support
MCP protocol
Open-source client

Simple pricing

Import all your history on signup. Your AI gets smarter the longer it remembers.

Free imports your full history on onboarding — no limits on the initial load. After that, free tier retains the last 30 days of active memory.

Free

$0
forever
  • 100K records
  • 30-day retention
  • Full history import
  • 500 queries/day
  • 1 project
  • Encrypted search
  • Insights & analytics
  • Cross-session intelligence
Start Free
POPULAR

Pro

$29/mo
or $290/yr (save 17%)
  • 2M records
  • Unlimited retention
  • Full history import
  • 50K queries/day
  • 5 projects
  • Encrypted search
  • Usage insights & analytics
  • Cross-session intelligence
  • Agent optimization tips
  • Email support
Start Trial

Team

$199/mo
per team, up to 10 seats
  • 10M records
  • Unlimited retention
  • Unlimited queries
  • 20 projects
  • Everything in Pro
  • Shared team memory
  • SSO (SAML)
  • Audit logs
  • Priority support
Start Trial

Enterprise

Custom
for large teams & compliance
  • Unlimited everything
  • VPC / private deploy
  • Custom SLAs
  • Security review
  • Dedicated support
  • SOC 2 (roadmap)
Talk to Us

FAQ

How does zero-knowledge encrypted search work?

CAPRISE applies a distance-preserving transformation to your vectors before upload. The server can find nearest neighbors without ever decrypting the vectors or content. Your encryption key stays client-side.

Which AI tools does Engram support?

Claude, ChatGPT, OpenClaw, Cursor, Claude Code, and Codex out of the box. The MCP server and REST API work with any agent or editor.

How much can I save on tokens?

Typical reductions are 50–80% by replacing static context dumps with targeted retrieval. Instead of loading 9,700 tokens of workspace files, Engram retrieves ~800 tokens of relevant context.

How fast is retrieval?

Median recall is 12ms via compressed HNSW vectors with entity chains. Optimized for interactive use — your agent doesn't wait.

Is the client open source?

Yes. The client-side encryption code is fully open source. Audit the encryption boundary yourself — trust, but verify.

Can I self-host?

Enterprise plan includes VPC and private deployment options. The open-source client works with any compatible vector store.

Give your AI a memory
it actually deserves

Free for up to 100K records. No credit card required.

curl -sL engram.ai/install | sh