After months of experimenting with every major AI model available through OpenRouter, I have just published my first public GitHub repository: opencode-seo. It is an AI-native SEO platform built specifically for OpenCode, and it is the most complete autonomous SEO toolkit I have ever built.
This is not a side project. It is the direct result of a systematic process of testing every frontier model on the market — GPT-4o, Claude Sonnet, Gemini, Mistral, DeepSeek — routing them through OpenRouter.ai to run identical SEO tasks side by side. The conclusion was unambiguous: OpenCode.ai was the superior environment for this kind of work. Not because of the model itself, but because of the way OpenCode handles agentic execution, file system access, and multi-step tool use. It is built for exactly this.
Why I Built It
Every SEO consultant I know has the same problem. The commercial tools — your Ahrefs, your Semrush, your Screaming Frog — are excellent crawlers. They are not intelligent. They hand you data and expect you to do the thinking. You still spend four to eight hours on a single site audit, writing up findings, building a report, and drafting a remediation plan that a client will read once and file away.
I wanted to collapse that process. Not by cutting corners, but by building a system that does the analysis, the prioritisation, the fix generation, and the client-ready report in one continuous, evidence-backed workflow. The entire suite runs on live DataForSEO data — every volume figure, every ranking, every backlink metric is pulled in real time. The system is explicitly instructed to refuse to fabricate a single metric.
What the Suite Actually Does
The repository ships with 82 skills, 4 specialist agents, 8 slash commands, and a deterministic rule engine containing 26 SEO checks written as YAML rules with embedded tests. That rule engine is the part I am most proud of. It runs in pure Python with zero model calls, which means the results are identical regardless of which of the 400+ models available on OpenRouter you are running. You can swap the underlying LLM entirely and the technical audit output does not change.
The architecture is built in three layers. Atomic skills handle focused, single-purpose tasks — checking canonical tags, generating schema, analysing a specific URL. Workflow skills chain those atomic skills together for end-to-end jobs: a full site audit, a content refresh cycle, a migration review. Project memory — stored in a simple seo-project.yml file or a per-client profile — keeps every output consistent across sessions. The agent knows your brand voice, your competitors, your target audience, and your technical constraints before it runs a single command.
The OpenRouter Experiment
Before settling on OpenCode, I ran the same suite of tasks across multiple models via OpenRouter. The experiment was not about which model produced the best prose. It was about which environment handled the full agentic loop without breaking — file reads, multi-step tool use, long-context retention across a 30-step audit, and deterministic code output for the fix engine.
Some models were faster. Some were cheaper. Some produced more elegant prose in the report writer. But when it came to the complete workflow — ingesting a live site, running 26 deterministic checks, generating mechanical fixes with .bak backups, and assembling a white-label HTML report with charts — OpenCode was the only environment that handled the entire pipeline without requiring manual intervention at any step.
That is the reason the suite is built for OpenCode specifically, rather than being a generic MCP server or a Claude Code extension.
AI Search Is a First-Class Citizen
One of the things that separates this suite from every commercial SEO tool on the market is the AI search layer. Ten dedicated skills cover AI Overviews, AI Mode, ChatGPT citation readiness, Perplexity optimisation, Gemini visibility, and LLM citation scoring. These are not bolt-ons. They are core to the architecture.
If you are still treating LLMO (Large Language Model Optimisation) as a future consideration rather than a present priority, you are already behind. The citation scorer runs an 11-criterion evaluation against every page you audit. It tells you exactly why a given page is or is not being cited by AI platforms, and it generates the specific structural and content changes required to fix it.
What It Costs to Run
This is where the comparison with commercial tools becomes stark. The suite is MIT licensed and free. The only ongoing cost is your DataForSEO API spend, which runs to pennies per audit rather than hundreds of pounds per month. A full site audit that would take four to eight hours manually and cost a significant slice of a senior analyst's day runs in ten to fifteen minutes.
For freelance consultants, the economics are transformative. You run a full audit on the discovery call, deliver a branded HTML report with charts and an executive one-pager the same afternoon, and the DataForSEO cost ledger keeps your spend visible per client so you can account for it accurately.
Get Started
The repository is live at github.com/venomous2/opencode-seo. Installation takes under five minutes on macOS, Linux, or Windows. The INSTALL.md walks through the full setup, including DataForSEO credentials and the optional Google API tier for Search Console, GA4, and CrUX field data.
If you want to discuss how to integrate this into your agency workflow, or if you are an enterprise team looking to deploy a repeatable audit cadence across a client portfolio, let's talk.