MarkovoFiles in. Clean Markdown out.

CompanyLast updated

About Markovo

Markovo is a file-to-Markdown pipeline for people and agents. You give it a PDF, Office document, image, audio or video file, or a public URL; it returns a clean Markdown file plus a reviewable bundle — extracted assets, a source map back to the original page regions, and a per-page quality report.

The product runs one account across every surface: the web converter on this site, a REST API, the markovo CLI, and an MCP server your assistant can call directly. Conversions are page-aware: digital text, tables, formulas, and scanned regions each take the route that fits them, and pages that need a human check are flagged instead of silently approximated.

What's open source — and what isn't

People regularly ask about the three repos, so here is the honest split:

fisher-byte/markovo · open

The client-side distribution: the markovo CLI and the markovo-mcp stdio MCP server — both ship from this one repo; there is no separate markovo-mcp repository. Install docs, MCP configuration, client source. As the repo's own README states: it contains no local conversion engine, backend service, deployment configuration, production credentials, or billing implementation — the client talks to the metered service at markovo.net.

fisher-byte/markovo-releases · open

Public release artifacts — pinned CLI wheels and the local Reader runtime packages, with sha256 manifests for reproducible installs. Both are client-side packages: the CLI drives the API, and the Reader runtime renders finished result bundles in your browser — neither contains conversion code, so nothing engine-related ships to your machine.

The conversion engine · not open

The pipeline that reads your pages — layout analysis, OCR routing, quality scoring — runs server-side at markovo.net and is closed-source, account-metered IP. There is no local or offline conversion path today — every conversion is a server job; that's a deliberate boundary, not an oversight. The repos are scoped so that what's public (clients, packaging, docs) carries no credential or engine leak risk. If local conversion matters to you, tell us — it's a real product question, not a secret.

Short version: markovo = how you install and drive the client (CLI + MCP, one repo); markovo-releases = where the signed client packages live; the engine = closed, runs on our side — the client packages can't convert anything by themselves.

Evidence, not vibes

Every completed job ships a bundle you can audit: output.md, extracted assets, source_map.json, quality_report.json, and a checksummed manifest. We run public benchmarks — the numbers, method, and limits are published here — and the homepage shows unedited conversion proofs with downloadable bundles.

Contact