Introduction
geopilot is an open-source, local-first, framework-agnostic technical readiness auditor and source-aware remediation assistant for websites.
Technical readiness only. GEO does not measure or predict citations, rankings, traffic or inclusion in any AI or search product.
Your coding agent audits your code on any framework with no install, explains what it finds in plain words, and fixes it with your approval. A deterministic CLI measures your running site when you want proof. The promise is narrow on purpose: find the observable issue, show the evidence, trace its source when reliable, propose the smallest safe change, and honestly report what was verified.
Two experiences, neither depends on the other
| Experience | What it can do | What it cannot claim |
|---|---|---|
| CLI without AI | Crawl sites on any stack; run deterministic rules; explain them; assess coverage; trace source on supported frameworks; apply approved deterministic patches; verify outcomes. | Understand arbitrary business intent, or guarantee search/AI performance. |
| Agent with your LLM | Consume CLI evidence; inspect a repository on any framework; explain context; suggest and, after approval, make contextual or editorial edits; re-run checks. | Turn its opinions into engine findings, or claim provenance or runtime verification the engine did not establish. |
What this release contains
- The universal URL audit (T0) for every rule that does not need a browser renderer.
- Next.js App Router and static HTML source mapping (T2) with the five closed-list fixes (T3): S1, S2, G1, G2, G3.
- Readiness, coverage, confidence and assessment scope, computed exactly as the worked examples show.
- V1–V4 verification, including the GEO-started serve process that is the only way source freshness is ever “confirmed”.
- Agent plugins for Claude Code, GitHub Copilot CLI, OpenCode and OpenAI Codex; insight validation (L1/L2) and report merging.
Not in this release
- A browser renderer: GEO-003 and GEO-002 report
not_evaluated(renderer_unavailable), never a failure. - Diff-aware
review, baselines, pre-push hooks, SARIF and the GitHub Action (Phase 3). - Crawler-specific robots fixes (G5) and app-link file creation (G4), which need the Phase 2 equivalence proof.
- A package for Cursor; the
AGENTS.mdsection is a provisional fallback.
Start with Getting started.