Can you patent an API or algorithm?
Short answer: Pure math on paper is not patentable; a specific computer implementation that applies an algorithm to solve a technical problem in a non-routine way may be.
Backend engineers patent routing logic, compression schemes, scheduling heuristics, and ML inference pipelines — but examiners frequently classify bare algorithms as abstract mathematical concepts. The difference is whether your disclosure and claims tie the algorithm to a particular machine architecture with a measurable technical effect.
API endpoints alone rarely qualify
“Receive JSON, validate fields, query database, return JSON” describes millions of REST services. Patentable subject matter, if any, lives in how validation, caching, batching, or federation works — specific data structures, ordering guarantees, or failure modes you invented.
Algorithm claims that struggle
- Steps a human could perform mentally or on paper.
- Standard ML inference (“train model, predict label”) without architectural novelty.
- Well-known graph algorithms applied to a new business domain only.
Algorithm claims with a stronger story
- Adaptive chunking or encoding that cuts bandwidth with defined invariants.
- Scheduling that co-locates tasks based on hardware telemetry you measure.
- Hybrid retrieval + scoring pipelines with non-obvious fusion steps tied to latency budgets.
Enablement (§112) for code-heavy inventions
Examiners expect enough detail that a skilled developer could implement the core idea without undue experimentation. Include pseudocode or representative modules, edge cases, and parameter ranges — not just marketing diagrams.
Next steps
Run a free score on the module that embodies your algorithm. If novelty and eligibility pillars are weak, iterate in the Interactive Code Review before engaging counsel. Read prior art search for code for GitHub and patent overlap patterns.