MAEL architecture overview

New to the MAEL dashboard? Start here.

A complete, beginner-friendly, step-by-step tutorial — no technical knowledge required. From your first login to your first published, AI-written article.

What is MAEL?

MAEL is an AI operating system purpose-built for SEO work: keyword research, content generation, technical audits, publishing, and monitoring — run by specialized agents, coordinated by a durable workflow engine, and gated by real evaluation before anything reaches a client’s site. It isn’t a prompt wrapper. Every run goes through one execution spine (LifecycleManager), every workflow step is idempotent under retry (ADR-006), every tenant’s data is isolated at the database role level (Row-Level Security), and every piece of generated content can be gated by a real quality evaluator before it publishes (content evaluation gate) — never a silent auto-pass.

Quickstart

Get MAEL running locally in under 10 minutes.

Architecture

How agents, memory, and the workflow engine fit together.

API Reference

Every endpoint, typed and documented from the live OpenAPI schema.

Deployment

Docker Compose, Kubernetes, or a single VPS — pick your path.

Why MAEL

Research, planning, content, technical SEO, publishing, monitoring, and authority agents all run through the same lifecycle envelope — retries, idempotency, cost tracking, and evaluation are handled once, centrally, not reimplemented per agent. See AI Agents.
DAG-based workflows with per-step checkpointing, heartbeat-based stale-run recovery, and compensation on failure. A worker dying mid-run does not lose the run. See Workflow Engine.
PostgreSQL Row-Level Security enforced at the database role level, with a fail-closed sentinel tenant — a session that forgets to set its tenant context sees zero rows, not someone else’s. See Multi-Tenancy.
Content generation runs through an evaluation gate — SEO, E-E-A-T, readability, hallucination risk, and (for medical/YMYL content) fact accuracy — before it’s eligible for human approval. An agent with no evaluator assigned is recorded as not evaluated, never scored 100 by default. See ADR-009.
Anthropic, OpenAI, Google, Ollama, and vLLM all load through the same provider registry with per-provider circuit breakers and automatic failover. Swap models per agent without touching code. See SDK.

Who this is for

MAEL runs SEO programs end to end: keyword research and clustering, AI-assisted content generation with a mandatory human-approval gate, technical SEO audits, publishing to WordPress, and ongoing rank/backlink monitoring — for agencies and in-house teams managing multiple client domains from one platform.
MAEL is currently in its production-validation phase — proving the platform end to end on real client domains before any self-serve commercial launch. See the Roadmap for what’s shipped and what’s next.

Get help

Troubleshooting

FAQ

GitHub