It reads your own Search Console, finds what to fix and what to write, drafts pages as pull requests for you to approve, and drips them out on a schedule. No invented data. No surprises.
Get started in 15 minutes See the dashboard (sample data) View on GitHubEvery step that changes your site waits for you. The rest runs on its own, every day, on GitHub's servers.
Daily Search Console report and an index check of your pages.
Keyword gaps from your real queries, plus news from the feeds you trust.
A "Topic" issue arrives with the evidence. Add the label approved, or close it.
Claude writes the page on a branch and opens a pull request. Your build checks it.
You read it and merge. Only then is the page queued.
Queued pages get dates, three a day, and the site rebuilds the day each is due.
Each job is separate. Turn on what you want. Then see the expert layer and the agent team.
Clicks, impressions, position, and which pages Google has indexed. Near-miss queries and pages with views but no clicks.
Weekly, from your own search data: near-miss, question, rising and uncovered queries. No fake volume numbers.
Reads RSS and Atom feeds you choose (regulators, platforms, your industry) and proposes at most one topic a day.
What changed, what it probably means, what to do next. Includes Google Analytics sessions when connected.
Pages arrive as pull requests with sources cited. The agent can only edit your content folders and cannot run code.
Approved pages are dated a few a day and released on schedule, so the site grows steadily.
Google cannot be automated, so you get the ten pages that need it, each with the full URL to paste into Search Console.
At most five free-listing attempts a day, one retry, everything logged. Stops at any bot check. Also finds unlinked brand mentions.
One HTML file per site, plus an app on your computer that manages every project in one place.
Every finding says why it matters and how to fix it, and every estimate says how far to trust it. The demo shows each of these with sample data.
Crawls your live site like a search engine: status codes, redirect chains, canonicals, duplicates, structured data (required properties), hreflang, Open Graph, orphan pages and broken links.
Queries stuck just off page one, pages that under-perform the click-through curve of your own site, pages competing for one query, decaying pages, intent and brand split, and topic gaps.
Everything scored as impact x confidence / effort, with the evidence next to it. A weekly issue lists the top ten.
Which AI search and training crawlers your robots.txt lets in, llms.txt, whether pages open with a direct answer, serve their text in the initial HTML, and carry entity and trust signals.
Lab and real-user speed for your top pages through Google's PageSpeed Insights, with the biggest fixes listed.
Log a change and get a before and after read, adjusted against your untouched pages. Labelled directional, because it is.
Backed by twenty-six written playbooks that say where the guidance comes from and where it is uncertain. See it in the demo.
Each agent has a fixed set of tools, required reading, a quality bar, a list of things it must never do and a defined output. You can read every word on the agents page, and use them in Claude Code too.
Turns all the data into at most five actions for the week, each with evidence, owner, effort and a stated confidence, plus what it is deliberately not doing.
Checks the site already covers a topic, reads the live results, decides the intent and writes the brief. Can recommend "improve the existing page" or "skip".
Writes one new page in your exact format, marked queued, with sources cited, and opens a pull request. Content files only.
Diagnoses why a page lost visibility first. Edits only when content is the cause, and makes the smallest change at the same URL.
Reviews every draft against a blocking checklist: unsupported claims, bad sources, wrong files, format breakage, duplicate targets. Optional.
Fixes audit findings at the cause using written recipes, then re-runs the audit. Runs locally with you watching, never unattended.
Checks every number, date and source on a page against a real source and your verified facts, and blocks what it cannot support.
Rewrites titles, descriptions and opening answers for pages that get impressions but few clicks, one change at a time, so it can be measured.
Finds pages competing for one query and plans a merge, redirect or re-target, with the exact redirect map.
Decides keep, refresh, merge, redirect or remove for thin and stale pages, with the evidence for each.
Adds structured data generated from the same content as the page, so the markup always matches what visitors see.
Keeps old URLs reaching the right page in one hop, and flags any change that silently removes a live URL.
Turns speed and Core Web Vitals findings into specific, measured code fixes, biggest cause first.
Tracks what Google has and has not indexed, explains why, and keeps the request-indexing list honest.
Finds genuine link opportunities and drafts outreach for you to send. Never sends anything, never recommends paid links.
Tests real questions in your field, records who AI answers cite, and fixes the gaps with clear, answer-first content.
Business profile, listings consistency, local markup and genuine location pages, for businesses that serve a place.
Reads the site profile and the agent plan, calls the right specialist in the right order and checks what comes back. One conversation, the whole team.
Profiles any platform, decides whether agents may edit files or only advise, and fills in the rules for the site.
Shows what the pages above you do better and where you can add something real, without copying anyone.
Language and country targeting: hreflang, canonicals, translation checks.
Categories, products, filters, stock and product markup that matches the page.
Plans a platform, domain or URL move so visibility survives it.
Makes the business a clear, consistent, trusted entity, and watches branded search and reviews.
Images, video and documents: accurate alt text, light files, crawlable, good previews.
Spots spam-policy, sensitive-topic and disclosure risks before they cost you.
Writes the plain-English report and judges honestly whether past changes worked, with the uncertainty stated.
Everything works with Node alone, with no dependencies. Two optional engines add speed and real agents, and are used only when they are installed.
Jobs, audit rules, priorities, the dashboard, the local app and the workflows. Zero dependencies, runs anywhere Node 20 does.
A concurrent, polite crawler for big sites with the same output as the built-in one. A test crawls the same site with both and requires identical findings.
Query topics, unusual days and trends from your own data, plus a tool-using agent runtime on the Claude Agent SDK. The agents read, fetch public pages safely and propose. They cannot edit or publish.
How the agents are kept safe, and what each engine does: engines documentation.
The tools that mass-produce pages put your rankings at risk. This one is designed the other way round.
So you do not find out the hard way.
Needs Node 20 and a GitHub repository. No account with us, no card, no signup.
# 1. scaffold config, data files and GitHub workflows npx github:bishal-paul-git/seo-autopilot#v2 init --url https://your-site.com/ # 2. connect Google, then check the setup npx github:bishal-paul-git/seo-autopilot#v2 doctor # 3. commit, push, and run Actions > seo-daily
One page on your computer for every website you run: charts, index status, schedule, keywords, backlinks, and the approvals waiting on GitHub.
npx github:bishal-paul-git/seo-autopilot#v2 projects add ~/sites/my-website npx github:bishal-paul-git/seo-autopilot#v2 ui --open
It listens only on your computer, needs a private link, and can only run dry runs and reports. See what the dashboard looks like.
MIT licensed. Install it in any repository and run it on GitHub's free tier. Documentation and a setup guide included.
Erudience, the company behind this, can set it up, connect Google and run it for your site.
No. It opens an issue with the evidence. A page is only drafted after you add the label approved, arrives as a pull request, and goes live only after it is merged and its scheduled day arrives. If you choose, you can let it merge drafts that pass your build and an optional reviewer agent, so approving the topic is your only step. That is off unless you switch it on.
No. The core runs on Node alone. Go speeds up the crawl on large sites, and Python adds the topic and anomaly analysis and the unattended agents. Each is detected automatically, and every job says plainly when something it needs is missing.
Only for drafting pages. Everything else (reports, audit, insights, priorities, AI search checks, speed, scheduling) runs without any AI key. For drafting, use your own Claude subscription token or an API key in your own repository. Nothing goes through us.
Reports, research, news, the digest, the indexing list and the dashboard work with any site that has a sitemap. Drafting and scheduling need pages stored as JSON or Markdown files in a repository. A CMS adapter would be a welcome contribution.
The software is free. GitHub Actions has a free tier that covers the daily jobs for a small site. AI keys and a cloud browser are optional and billed by their providers.
No. Search data goes from Google to your own GitHub runner and is saved in your own repository. Optional AI calls send short snippets to the provider you choose.
The drafting agent can only edit the content folders you list, cannot run build or test commands, treats issue text as data, and cannot push to your main branch. Your build runs separately with no secrets. See the security notes.