31ms Zero-Copy PostgreSQL 16 & pgvector Branching

1. Is Mnemon-DB a proprietary database engine or real PostgreSQL?

100% genuine, upstream PostgreSQL 16. Mnemon-DB is not an emulation layer, proxy translation, or custom SQL dialect. It runs standard PostgreSQL 16 binaries and supports all standard PostgreSQL extensions, including pgvector, pg_trgm, uuid-ossp, and postgis. Any application or ORM that speaks PostgreSQL (Prisma, Drizzle, SQLAlchemy, Hibernate, libpq, psycopg3) connects without modifications.

2. How does 31ms / 4ms database branching actually work? Does it duplicate data?

Zero-copy copy-on-write (reflink) file system virtualization. Mnemon-DB operates at the Linux file system block layer on NVMe storage (using XFS or Btrfs reflinks):

• Zero Data Duplication: When you branch a 500GB production cluster into a new tenant branch, zero data blocks are copied. The branch immediately points to the immutable base data blocks of the parent cluster, consuming 0 extra bytes of storage.
• Write Isolation: As the new branch executes SQL queries, updates, and inserts, only newly written or modified 8KB page blocks are written to disk.
• Branch Latency: Cold branch creation takes 31.6ms; warm branch instantiation on pre-warmed parent templates takes sub-4ms (typically 1.8ms).

3. How does Scale-to-Zero work for PostgreSQL, and what is actually free when idle?

Traditional managed databases (like AWS RDS or Aurora Serverless) consume constant CPU and RAM even when nobody is querying them.

Mnemon-DB's Memory Yield: When a tenant branch experiences zero queries for its idle window (default: 5 minutes), the engine issues a soft-freeze (via Linux cgroups v2 freezer or SIGSTOP), serializes connection states, and releases physical memory. Guest CPU utilization drops to 0.00%, and RAM is 100% reclaimed.

The Billing Guarantee: You pay $0.00 for idle database CPU and $0.00 for idle database RAM. You only pay for raw block storage of modified delta pages.

4. What is the blast radius and isolation between database branches?

Each branch runs with its own isolated PostgreSQL postmaster process, dedicated Write-Ahead Log (WAL) journal, separate network socket, and distinct credentials.

Total Mutation Isolation: If an autonomous AI agent or test runner executes destructive queries inside a branch (e.g. DROP TABLE users;, TRUNCATE TABLE transactions;, or corrupts schema definitions), the parent production cluster is 100% unaffected. The branch can be discarded and recreated in 4 milliseconds.

5. How does pgvector support work for AI agent memory and retrieval?

Mnemon-DB comes with pre-compiled pgvector 0.7.0+ supporting high-dimensional vector embeddings (e.g. vector(1536) for OpenAI or vector(3072) for Claude) alongside HNSW and IVFFlat index types.

Branching Vector Stores: AI coding agents can branch production vector stores in 31ms, test speculative indexing or fine-tuned embedding modifications, evaluate RAG retrieval accuracy in parallel, and commit or discard without touching the live vector index.

6. How does Mnemon-DB compare to AWS Aurora Serverless v2 and Neon?

• vs. AWS Aurora Serverless v2: Aurora requires a 0.5 ACU minimum ($43.80/month per idle instance) and takes 15–20 minutes to clone a database cluster. Mnemon-DB scales to true $0.00 idle and branches in 31ms.
• vs. Neon: Neon relies on a complex, proprietary Rust Pageserver architecture and external cloud control plane. Mnemon-DB is self-hostable on your own hardware or AWS VPC with zero external cloud dependencies and 100% data sovereignty.

7. How do I remotely connect, branch a database, and run SQL queries?

You can interact with Mnemon-DB via REST API or direct PostgreSQL wire protocol:

A. Instant Database Branching (cURL):

curl -X POST "https://ciel.theeducode.com/v1/tenants/branch" \
  -H "Authorization: Bearer esk_live_demo10k" \
  -H "Content-Type: application/json" \
  -d '{
    "parent_cluster": "main",
    "tenant_id": "agent-test-branch-01",
    "enable_pgvector": true
  }'

psql "postgres://ciel_admin:secret@ciel.theeducode.com:5432/agent-test-branch-01?sslmode=require"

from ciel import MnemonClient

client = MnemonClient(api_key="esk_live_demo10k")
# Branch 500GB production database in 31ms
branch = client.branch(parent="production", branch_name="pr-1284-test")
conn = branch.get_connection()
cursor = conn.cursor()
cursor.execute("SELECT COUNT(*) FROM users")
print(f"Users in branch: {cursor.fetchone()[0]}")
branch.destroy()  # Clean up ephemeral delta in 2ms