Your code contains inventions you don't know about yet.
Obviously Not reads your codebase and surfaces the distinctive technical concepts inside it, each one scored and tied to the exact code that implements it. Free and open source. Run it on your own machine, in CI, or inside your own AI agent.
👉 Scan Your CodeCode scanners ask what is wrong. We ask what is distinctive.
Not a patent scanner. It reads your code for distinctive technical concepts. That is engineering work, not legal work. Whether any concept is worth patenting is a separate question for a patent attorney, and that is the only place the legal judgment belongs.
Distinctive engineering rarely announces itself.
def uncertainty_guided_cache_invalidation(query, E, cache):
"""ML-guided selective invalidation using uncertainty/drift"""
drift = cosine_sim(E[query], cache.E).std() # semantic drift (uncertainty)
thresh = base_thresh * (1 + cache.staleness(query)) # adapt to recency/frequency
if drift > thresh:
keys = cache.semantic_neighbors(query, radius=drift) # target related entries only
return cache.invalidate(keys, decay=drift) # selective, proportional purge
return cache.get(query)
Nobody writes a design doc for a loop like this. It ships, it works, and the reasoning behind it lives in one engineer's head. That is the material the scanner is built to find and write down.
See it on a codebase you can check.
We scanned PX4 Autopilot, a mature open-source flight stack, and published the results in full. The repository is public, so you can read the findings against the actual source.
How it works.
Pick how you want to run it
Download the CLI and scan locally with Ollama, add the GitHub Action to your pipeline, or install the agent skills into Claude Code, Gemini CLI, or Cursor. Same engine, three entry points. All open source, all free.
Point it at your code
It reads your source the way an engineer would, looking for the decisions that are not standard practice. With the local CLI, nothing leaves your machine: no cloud upload, no API key, no external call.
Review what comes back
Each concept arrives with a plain-language description, the problem it solves, scores across four engineering axes, and the specific files and line ranges behind it. Textbook findings are held back by default, but the count is always reported. Nothing is silently dropped.
Built for builders.
The distinctive work is already in your repository. The problem is that it was never written down anywhere a reviewer, an investor, or an attorney could read it.
Software Engineers
The interesting decisions are in the failure handling, the routing, the thing you tried three ways before it worked. That reasoning lives in your head and nowhere else.
Scan your codeTech Founders
Diligence asks what is defensible about your technology. A source-linked inventory of your distinctive engineering is a better answer than a slide.
Build the inventoryVibe Coders
Building fast with AI means shipping systems nobody documented, including you. The work is still yours. Finding out what you built is the first step.
See what you've builtAnd patent attorneys.
Clients cannot disclose what they never noticed they built. The scanner reads the codebase directly and hands you a structured technical description with the source evidence attached.
Intake that starts from code
Scored concepts with file and line-range citations, instead of a blank disclosure form and an hour of interview to fill it in.
Anchored to implementation
Every finding is tied to the specific mechanism in the source and the problem it solves, which is the level of technical detail an eligibility analysis needs to work from.
Scans where the work happens
Repositories and pull requests, not Confluence pages and meeting notes.
The legal call stays yours
We find and structure the engineering. Novelty, eligibility, inventorship, and what to claim are yours. The tool is built around that line, not apologetic about it.
Managing a patent portfolio?
Enterprise patent teams need systematic invention surfacing across engineering orgs. Most rely on engineers self-reporting, which means most innovations go unprotected.
Systematic discovery
Scan across repositories and teams to surface the work engineers never flag. Portfolio building stops depending on which engineers happen to volunteer.
Scales with your org
From a single team to hundreds of engineers. Structured disclosures feed directly into your existing patent workflow.
Runs inside your perimeter
Local-first scanning keeps source code within your corporate environment: no external API calls and no cloud uploads. Security, export-control and client-policy requirements still apply to the particular deployment.
We publish the work, not just the claims.
This is a field full of assertions nobody can check. We would rather show the measurements and let you disagree with them.
What local AI actually costs
We scored 29 models across ten specifications with a four-vendor panel to measure what self-hosted drafting gives up. Open weights cost about 4 points. Limited memory costs about 23.
Read the studyWho actually gets named
1.68 million US patents merged with EEO-1 workforce data to rank 91 tech companies on how technical work becomes inventorship. At most firms the top 5% of inventors file most of the patents.
See the indexThe reference library
Case law, disclosure templates, and engineering guides on inventorship, eligibility, and disclosure timing. Free, sourced, and written for people who read code.
Browse resourcesYou've already built something distinctive. Find out what it is.
Local CLI, GitHub Action, or agent skills. Open source and free. With the local CLI your code never leaves your machine.
The same scanning packaged as open-source skills for Claude Code, Gemini CLI, and Cursor. Portable, MIT-0, and readable line by line before you run it.
Want to learn more about protecting your innovations? Check out our Resources →