# Open SEO Key Features: 4 Core Capabilities That Power Serverless SEO

> Discover Open SEOs core features: AI Onboarding Agent, scalable Site-Audit Crawl, SQLite state management, and a deterministic LLM pipeline. Power serverless SEO affordably on Cloudflare Workers.

- Repository: [Every App/open-seo](https://github.com/every-app/open-seo)
- Tags: deep-dive
- Published: 2026-08-13

---

**Open SEO's key features include an AI Onboarding Agent, scalable Site-Audit Crawl, SQLite-backed state management via Durable Objects, and a deterministic LLM pipeline—all running serverless on Cloudflare Workers for pennies per user.**

Open SEO is a serverless SEO platform built entirely on Cloudflare's edge infrastructure. Understanding the **open SEO key features** requires examining how its architecture delivers enterprise-grade SEO tooling without traditional server costs. This article breaks down the four foundational capabilities implemented in the codebase.

## Onboarding Agent: Free SEO Strategy Generation

The **Onboarding Agent** is Open SEO's primary user acquisition and activation mechanism. It provides a guided chat experience that delivers personalized SEO strategies at minimal cost.

### How the Pipeline Works

The agent follows a deterministic three-stage pipeline:

1. **Live site scraping** – reads the user's actual website content
2. **Keyword research** – fetches relevant keyword ideas from DataForSEO
3. **LLM synthesis** – streams a complete SEO plan back to the user

This entire flow costs only a few cents per signup, making it viable as a free offering. The implementation specification lives in [[`specs/0005-onboarding-agent.md`](https://github.com/every-app/open-seo/blob/main/specs/0005-onboarding-agent.md)](https://github.com/every-app/open-seo/blob/main/specs/0005-onboarding-agent.md).

### Technical Implementation

The agent leverages Cloudflare Workers' edge location to minimize latency during the streaming response. By keeping the pipeline deterministic—fixed scrape → keyword → LLM ordering—Open SEO ensures reproducible outputs and predictable resource consumption.

## Site-Audit Crawl: Fault-Tolerant Web Crawling

Open SEO's **Site-Audit Crawl** provides comprehensive website analysis through a scalable, resilient architecture.

### Cloudflare Workflow Architecture

The crawl engine runs as a **Cloudflare Workflow**, which automatically handles:

- **Parallel execution** – distributes crawl tasks across multiple Workers
- **Fault tolerance** – retries failed requests with exponential backoff
- **Resource limits** – respects rate limits and prevents runaway compute

### SQLite-Backed State Management

All crawl state persists in **Durable Objects** with SQLite storage:

| State Type | Storage Purpose |
|------------|---------------|
| Crawl frontier | Queue of URLs pending discovery |
| Link graph | Discovered internal/external link relationships |
| Page mirrors | Cached page content for analysis |

This design eliminates external database dependencies while maintaining ACID guarantees for crawl consistency.

## Durable Objects: Edge-Native Persistence

Open SEO's use of **Cloudflare Durable Objects** is foundational to its serverless architecture. Unlike traditional serverless platforms that require external databases, Open SEO co-locates compute and state at the edge.

### SQLite Integration

The platform uses **SQLite-backed Durable Objects** introduced in Cloudflare's 2023 runtime updates. Each Durable Object instance:

- Runs in a single JavaScript isolate with exclusive SQLite access
- Supports 1GB+ of storage per object
- Provides synchronous SQL execution via the `D1Database` API

This enables complex relational queries for link graph analysis without network round-trips to external databases.

## Deterministic LLM Pipeline: Cost-Controlled AI

The final **open SEO key feature** is its rigorous approach to **deterministic LLM execution**. Rather than open-ended agent loops, Open SEO structures all AI interactions as fixed pipelines.

### Pipeline Characteristics

- **Fixed execution order** – no tool-calling loops or recursive planning
- **Token budgeting** – hard limits on per-request and per-user consumption
- **Streaming responses** – progressive rendering reduces time-to-first-byte

This architecture choice directly enables the platform's economics: sub-dollar costs for complete onboarding sessions that would cost dollars on conventional agent frameworks.

## Summary

Open SEO delivers a complete SEO platform through four tightly integrated capabilities:

- **Onboarding Agent** – deterministic chat pipeline generating free SEO strategies via DataForSEO and LLM synthesis
- **Site-Audit Crawl** – fault-tolerant web crawler running as Cloudflare Workflows with SQLite state persistence
- **Durable Objects architecture** – edge-native storage eliminating external database dependencies
- **Deterministic LLM execution** – fixed pipelines ensuring predictable costs and reproducible outputs

## Frequently Asked Questions

### How does Open SEO keep costs so low per user?

Open SEO minimizes costs through deterministic LLM pipelines with no recursive tool calling, edge-native SQLite storage avoiding external database fees, and Cloudflare Workers' generous free tier. The onboarding agent specification explicitly targets "a few cents per signup" by eliminating agent loop overhead common in frameworks like LangChain or AutoGPT.

### What database does Open SEO use for crawl state?

Open SEO uses **SQLite-backed Durable Objects** for all persistent state. Each crawl job receives its own Durable Object instance with isolated SQLite storage for the frontier, link graph, and page mirrors. This co-locates compute and storage at Cloudflare's edge locations.

### Why is the Onboarding Agent deterministic rather than autonomous?

The deterministic design—fixed scrape → keyword → LLM ordering—ensures **predictable costs** and **reproducible outputs**. Autonomous agents with tool-calling loops can generate unpredictable token consumption, making free-tier offerings economically unviable. The specification at [`specs/0005-onboarding-agent.md`](https://github.com/every-app/open-seo/blob/main/specs/0005-onboarding-agent.md) mandates this constraint explicitly.

### Can the Site-Audit Crawl handle large websites?

Yes. The Cloudflare Workflow architecture provides automatic **horizontal scaling** across thousands of URLs. State persistence in Durable Objects means crawl jobs survive Worker evictions and resume gracefully. The SQLite backend in each Durable Object supports link graphs with millions of edges.