We teach patents
to reason.
Built for humans. Legible to agents.
Where this began.
[One paragraph — the trigger moment. What were you working on when you realised existing tools stop at the claim boundary? A specific patent search that failed? A research project that kept hitting the same wall? A client review that took three weeks when it should have taken three days?]
[Second paragraph — how the two of you came together. Where you met, the first prototype you shipped, the decision to make this your full-time bet. Keep it honest and specific. Two sentences is enough.]
By early 2026 we had joined Campus Founders in Heilbronn as a two-person GbR, with a working decomposition pipeline running nightly across the US, European, and Japanese patent offices.
A universal cognitive layer for IP data.
Your tools stop at the claims. We start from there.
Patents are the largest structured record of human invention, and the worst-formatted one. A single claim is a century-old paragraph of dense legal prose; the entities, dependencies, and functional relationships inside it are implicit — never labelled.
Agents can read those paragraphs. None can reason over them. That gap is the product.
IPable decomposes every claim into typed nodes — preambles, elements, limitations, functional language — resolves them across jurisdictions into canonical families, and serves the resulting graph through an MCP server for agents and the IPable platform for humans. One source of truth, two surfaces.
Three inflections met at the same moment.
Models went from reading to acting.
Tool use and protocols like MCP turned LLMs into agents that navigate real systems. They need structured interfaces — not more prose to summarise.
IP data keeps doubling.
140M+ documents today, ~3.5M new grants per year, split across a dozen jurisdictions with incompatible schemas. Manual review has run out of headroom.
Reasoning errors are existential.
Freedom-to-operate, invalidity, patentability — every downstream decision depends on getting the claim structure right. Hallucinations are not acceptable here.
Four things we won't compromise on.
Cite every answer back to source.
Every node in the graph carries a stable reference to the originating filing. If we say it, you can trace it to a USPTO paragraph number — no exceptions, no synthesised prose.
Deterministic where it matters.
Decomposition is a pipeline, not a prompt. The same claim in, the same graph out. Models live at the edges; the core is reproducible.
EU-first on data handling.
Hosted in Germany, processed in the EU, aligned to ISO/IEC 27001 and GDPR from day one. Customer data never trains a model, full stop.
Agent-native or nothing.
Every capability ships behind MCP and REST before it ships a button. If an agent can't use it in a tool-call loop, we haven't finished the feature.
Inside Gravity, Heilbronn.
We build from Gravity, the flagship startup floor at Campus Founders on the Bildungscampus — the same complex that hosts the TUM Heilbronn campus and a dense cluster of AI, industrial, and deep-tech teams backed by the Dieter Schwarz Foundation.
Being in the EU, hosted in the EU, and governed by German law isn't incidental — it's the whole pitch to our first market.
Gravity · Campus Founders
Bildungscampus 11
74076 Heilbronn, Germany
We're two.
Talk to us directly.
No BDRs, no forms that vanish into a CRM abyss. Whichever of us replies will be one of the two people actually building the product.