# mxto.ai

> mxto is the AI delivery engine for the Mendix™ platform. It reads any Mendix™ app into a typed, queryable representation an LLM can understand, authors every construct type, then compiles, deploys and tests it to a running app: AI-driven, headless, no Studio Pro. Read coverage and write coverage are both 100% (70/70 construct types certified), round-trip-proven.

mxto.ai is a product of Ontology Labs, Inc. (United States), powered by AYIOS.

Mendix™ is a trademark of Siemens Digital Industries Software. mxto.ai is not affiliated with, endorsed by, or sponsored by Mendix Technology B.V. or Siemens AG.

## Capabilities

- **Reads it**: Reads any Mendix™ app the way an LLM needs to: in order, by name. 94.4% fewer tokens than the raw model graph (17.9:1), 99.73% extraction coverage, 100% read coverage (70/70 construct types).
- **Writes it**: Authors every construct type (entities, microflows, pages, security) and proves it didn't break anything. 70/70 construct types certified, 0 unexpected diffs across 4,967 round-trip operations, 6,122 live model operations.
- **Ships it**: Compile → deploy → test → a running app, with Mendix's own build tools (mxbuild). MX 8 through 11, healthy in 20 seconds. Live debugger with semantic breakpoints.
- **Bring your own agent**: A Mendix™ model is a binary file a general AI agent can't open. mxto is the toolchain it drives: 142+ tools and a clean CLI built for an LLM. Bring Claude Code, Cursor, or your own agent.

## The benchmark

We measured what it costs an AI to read one Mendix™ microflow. A Mendix™ microflow is stored as a graph wired by opaque IDs. A general AI must resolve ~29 of them (2,592 tokens) before it understands one flow. mxto reads the same flow in order, by name: 145 tokens, 0 resolutions. 94.4% fewer tokens, ~29 → 0 inference steps. (Measured with tiktoken o200k_base on a real flow; the per-flow model graph is a conservative, baseline-favourable comparison.)

## Proof

- 70/70 construct types certified: read 100% + write 100%, 0 non-automatable
- 0 unexpected diffs across 4,967 round-trip operations on a real production estate
- 48,385 logic expressions parsed at a 100% parse rate
- 6,122 live model operations on a production application
- Mendix™ 8 through 11, healthy in 20 seconds
- Read-certified across 8 production Mendix™ estates with zero transform failures

## Case Studies

Anonymized, evidence-grounded accounts of mxto on real production Mendix™ applications. Every figure is measured on the client's own model. Index: https://mxto.ai/case-studies/. Each study has a machine-readable Markdown companion you can read directly.

- **CS-001: Reading a production lender's interest engine (comprehension + a latent-defect catch).** A South African consumer-lending platform's production Mendix™ app (77 modules, ~990 entities, ~3,400 microflows, 745 pages) read fully offline in 112 seconds, no Mendix™ SDK, with a 100% zero-difference round-trip. mxto produced an auditable specification of the interest-accrual engine and, in the read, surfaced a dormant leap-year century-rule defect (verified against the raw model, dormant until 2100). Page: https://mxto.ai/case-studies/cs-001 · Machine-readable: https://mxto.ai/case-studies/cs-001.txt

## Blog

The mxto.ai blog is written by an AI agent, and every post is machine-readable: alongside the human page there is a plain-text companion an AI can read directly. The product that makes Mendix™ apps AI-readable has an AI-authored, AI-readable blog. Every post passes the same gate before publishing: a non-slop review, a per-figure truth check against source, and a patent-safe and trademark pass. Then a human publishes. Index: https://mxto.ai/blog/

- **We read a production lender's whole app offline in 112 seconds, and found a dormant defect** (case study, written by AI). A South African consumer lender's production Mendix™ app (77 modules, ~990 entities, ~3,400 microflows, 745 pages) read fully offline in 112 seconds, no Mendix™ SDK, with a 100% zero-difference round-trip. The read produced an auditable specification of the interest-accrual engine and surfaced a dormant leap-year century-rule defect (verified against the raw model, dormant until 2100). The narrative companion to CS-001. Page: https://mxto.ai/blog/reading-a-production-lender · Machine-readable: https://mxto.ai/blog/reading-a-production-lender.txt
- **Round-trip or it didn't happen: proving an AI didn't drop anything** (fidelity, written by AI). The only honest proof that an AI read your whole app, or changed only what it meant to, is a round-trip: out to a representation, back to the model, compared. On a real production estate: equivalence 100%, traceability 100%, round-trip PASS at 4,212 constructs with 0 differences; and, as a published product figure, 0 unexpected diffs across 4,967 round-trip operations. Page: https://mxto.ai/blog/round-trip-or-it-didnt-happen · Machine-readable: https://mxto.ai/blog/round-trip-or-it-didnt-happen.txt
- **Why we let an AI write this blog, and how we keep it honest** (manifesto, written by AI). Why an AI-native product publishes an AI-written, machine-readable blog, and the editorial gate that keeps it honest. Page: https://mxto.ai/blog/why-an-ai-writes-this-blog · Machine-readable: https://mxto.ai/blog/why-an-ai-writes-this-blog.txt
- **How hard is it for an AI to read one Mendix™ microflow? We measured it.** (benchmark, written by AI). A Mendix™ microflow costs a general AI ~29 reference resolutions and 2,592 tokens to read; mxto reads the same flow in order, by name: 145 tokens, 0 resolutions (94.4% fewer, measured with tiktoken o200k_base). Page: https://mxto.ai/blog/reading-one-mendix-microflow · Machine-readable: https://mxto.ai/blog/reading-one-mendix-microflow.txt
- **Running any Mendix™ app in Docker (7–11) without bloating your disk** (engineering). Open-source Docker recipes that bake the Mendix™ runtime once per version and bind-mount the model (no binaries committed). Page: https://mxto.ai/blog/mendix-runtime-crates

## Proof pack

Downloadable, reproducible evidence behind the claims, anchored on Claudius (a complete Mendix™ app authored end-to-end by an AI agent driving mxto, which Ontology Labs owns and you can rebuild). Hub: https://mxto.ai/proof/

- Reproducible proof pack (one real app): extract, AI change, build, test, clean diff, with the readable flow, the typed IR, and the model graph downloadable. https://mxto.ai/proof/
- How mxto and Mendix™ Maia fit together (native IDE assistant alongside the external, headless delivery engine for estates, CI, and agents). https://mxto.ai/proof/vs-maia
- Security and enterprise readiness (local processing, the AI-agent data boundary, SOC 2 not certified by design). https://mxto.ai/proof/security
- What mxto does and doesn't do yet (honest boundaries; structurally safe is not the same as semantically correct). https://mxto.ai/proof/boundaries
- Inspectable proof files (Mendix™ versions tested, tools and capability map, system requirements, comprehension-benchmark method). https://mxto.ai/proof/files/

## The semantic layer

mxto makes AI-in-Mendix™ structurally safe: the app builds, the model round-trips, nothing drops silently. Making an app mean what you intend is the next layer: the semantic work done at Ontology Labs (https://ontologylabs.ai).

## Machine-Readable Discovery

- AI plugin manifest: https://mxto.ai/.well-known/ai-plugin.json
- Curated sitemap: https://mxto.ai/sitemap-ai.xml
- Full sitemap: https://mxto.ai/sitemap.xml

## Brand Family

mxto.ai is part of the Ontology Labs brand family:
- [Ontology Labs](https://ontologylabs.ai) - Parent company, Ontology Labs, Inc. (corporate, investment)
- [AYIOS](https://ayios.ai) - Parent platform (AI-native application stack)
- [mxto.ai](https://mxto.ai) - This product (the AI delivery engine for Mendix™)
- [Agileworks Group](https://agileworksgroup.ai) - Distribution partner
- [AWG Finance](https://moneyworks.ai) - Accounting software distribution

## Contact

- [Request a demo](mailto:hello@ontologylabs.ai?subject=mxto.ai%20demo%20request)
- [AYIOS](https://ayios.ai) - Parent platform
- [Ontology Labs](https://ontologylabs.ai) - Parent company