06 — Deployment

Local, zero dependencies

from memnex import Memnex
mx = await Memnex.create(tenant_id="demo")

Uses in-memory backends. Great for tests, scripts, notebooks.

Local stack (Docker)

docker compose up -d postgres redis qdrant      # backends only
export MEMNEX_POSTGRES_URL=postgresql://memnex:memnex@localhost:5432/memnex
export MEMNEX_REDIS_URL=redis://localhost:6379/0
export MEMNEX_QDRANT_URL=http://localhost:6333
export GOOGLE_API_KEY=...
memnex db init
memnex serve rest --port 8500

For the full stack (databases + the Memnex container itself), docker compose up -d.

Production checklist

Compute	Any container runtime. CPU-bound (minus DB). 1-2 vCPU per instance typical
Postgres	15+, pgcrypto enabled, RLS on, backups daily, connection pooler (PgBouncer)
Redis	7+, AOF persistence if you want warm caches across restarts, ACLs per tenant
Qdrant	Optional; one collection per tenant (default)
Auth	Set MEMNEX_API_KEYS. Front with a gateway for TLS
Rate limit	Swap the in-memory limiter for a Redis-backed one at scale
Observability	Scrape /api/v1/metrics. Dashboard in grafana/dashboards/
Workers	Run compaction + privacy workers separately from the API

Environment variables

Var	Purpose
MEMNEX_TENANT_ID	Default tenant for this process
MEMNEX_POSTGRES_URL	postgresql://...
MEMNEX_REDIS_URL	redis://...
MEMNEX_QDRANT_URL	http://...
MEMNEX_KAFKA_BROKERS	Optional, for event streaming
MEMNEX_LLM_PROVIDER	openai / anthropic / ollama / none
MEMNEX_LLM_MODEL	Model name
MEMNEX_LLM_API_KEY	Provider key
MEMNEX_API_KEYS	key1:tenant1,key2:tenant2

Scaling roadmap

Load	Change
< 1k QPS	Single process. In-memory limiter is fine
1k–10k QPS	Horizontal scale API. Move limiter to Redis. Enable Qdrant
> 10k QPS	Read replicas for Postgres. Shard by tenant across Postgres + Redis.

Upgrade strategy

• SQL migrations are idempotent; re-running memnex db init is safe.
• Config is a Pydantic model; new fields default safely.
• In-place rolling restarts work — no sticky state on any instance.