Fastest
Eureka on the web
When you want the answer in the next five minutes.
The agent works the prompt against patents and technical literature, citing every source.
Run your analysis now →https://www.patsnap.com/resources/blog/rd-blog/retrieval-augmented-generation-pipelines-patent-landscape/ · Patsnap · data cut-off 2026-07-31 · downloaded from the live page
Patent Landscape · Vector Search & RAGRetrieval Augmented Generation Pipeline Patents: Filing Trends and Claim Concentration
- Filings jumped from zero in 2017 to a peak of 261 in 2025, then a partial 2026 count of 35 that understates true filing activity given the roughly 18-month publication lag.
- The top 5 assignees hold 19.4% of all 489 records, and the top 10 combined reach 31.7% — real concentration, but not a closed field.
- G06F digital data processing touches 73.8% of records, while G16B bioinformatics and G06V image recognition sit under 5%, marking the thinner, less-claimed branches.
489
Published Records
19%
Top-5 Share of All Records
US
Leading Jurisdiction
100
Active Filers Ranked
Top-5 share is the combined record count of the five largest assignees divided by all 489 records in scope (CR5), not by the ranked leaders only.
Published byPatsnap Research··7 min readSourced from Patsnap Eureka
Overview
What this landscape covers
This review tracks 489 published patent records matching claim language around retrieval augmented generation pipelines: document chunking, chunk overlap, prompt construction, context window management, relevance ranking, reranking stages, citation attribution and index freshness. The search string pairs a RAG-pipeline term with a technical control term, so records without both are excluded — this is a claims-level view of the pipeline mechanics, not a broad AI-in-general set.
Filing activity is concentrated in the electric digital data processing class, with a substantial secondary cluster in AI-model computing classes. Publication lags filing by roughly 18 months, so the most recent year in any trend chart understates actual filing activity for that period.
Filing activity by year, 2017–2026
Source: Patsnap Eureka. Assignee ranking and totals. Derived from a Patsnap search on Retrieval Augmented Generation Pipelines covering 2015–2026, data cut-off 2026-07-31. Counts reflect published records only and shift as new filings publish.Run this in Eureka MCP
The Data
Filing trend and technology composition
The counts below come directly from the underlying record set: a filing-year trend across the full coverage window and an IPC subclass breakdown showing where RAG pipeline claims physically sit.
From zero to a 2025 peak
Filings were effectively absent through 2017 and climbed to a peak of 261 records in 2025. The 2026 figure of 35 is a partial-year count and should not be read as a slowdown — it reflects records not yet published rather than filings not yet made.
Where the claims sit
G06F (electric digital data processing) appears in 73.8% of the 489 records in scope, and G06N (AI-model computing) in 32.3% — the two anchor classes for pipeline mechanics. Business/admin (G06Q), transmission (H04L), speech (G10L), healthcare informatics (G16H), bioinformatics (G16B) and image recognition (G06V) each sit at 8.2% or below, since a record can carry multiple classes these shares sum to more than 100%.
Shares are the percentage of the 489 records in scope. A patent can carry several IPC classes, so the shares add up to more than 100%.
Source: Patsnap Eureka. Filing trend and technology composition. Derived from a Patsnap search on Retrieval Augmented Generation Pipelines covering 2015–2026, data cut-off 2026-07-31. Counts reflect published records only and shift as new filings publish.
Go deeper on Retrieval Augmented Generation Pipelines with Eureka
This page is one run against one query. Ask Eureka your own question about retrieval augmented generation pipelines and every answer comes back with the patent numbers behind it.
Try EurekaKey Patents
Representative filing and most-cited records
Representative Filing
Improving retrieval accuracy in a RAG framework
LTI MINDTREE LTD
The filing describes an adaptive pipeline that analyses input documents with a small language model to classify document type before extraction, then chunks content based on use case, latency, cost and the target large language model's context window size. Tokenization strategy and embedding model selection are likewise made variable, with quantization policies applied to produce the final vector representations.Filed by LTI Mindtree, published 2026-07-02 — illustrates how recent claims are moving from fixed chunking rules toward document-type-aware, resource-adaptive pipelines.
Most-cited records in this set
| # | Publication no. | Patent title | Citations |
|---|---|---|---|
| 1 | US20050060643A1 | Document similarity detection and classification system | 1,290 |
| 2 | US20240386015A1 | Composite symbolic and non-symbolic artificial intelligence system for advanced reasoning and semantic search | 370 |
| 3 | US20240412720A1 | Real-time contextually aware artificial intelligence (AI) assistant system and a method for providing a conte… | 231 |
| 4 | US20250259085A1 | Convergent Intelligence Fabric for Multi-Domain Orchestration of Distributed Agents with Hierarchical Memory … | 208 |
| 5 | US12417214B1 | System and method for adaptive semantic parsing and structured data transformation of digitized documents | 78 |
| 6 | US11861320B1 | Text reduction and analysis interface to a text generation modeling system | 63 |
| 7 | US20250259041A1 | Ai agent decision platform with deontic reasoning | 61 |
| 8 | US20250259043A1 | Platform for orchestrating fault-tolerant, security-enhanced networks of collaborative and negotiating agents… | 56 |
| 9 | US20250259075A1 | Advanced model management platform for optimizing and securing ai systems including large language models | 51 |
| 10 | US20250259082A1 | Ai agent decision platform with deontic reasoning and quantum-inspired token management | 47 |
Citation counts favour older records simply because they have had longer to accumulate citations inside the searched corpus — treat this as a signal of influence, not of current technical importance.
Each row carries its publication number; clicking a row searches Eureka by that number.
Source: Patsnap Eureka. Citation counts and representative records. Derived from a Patsnap search on Retrieval Augmented Generation Pipelines covering 2015–2026, data cut-off 2026-07-31. Counts reflect published records only and shift as new filings publish.Run this in Eureka MCP
Run it yourself
Put your own technology through the same analysis
Where to run it
For builders
MCP server & REST API
When it has to run inside your own pipeline.
Patent search, landscape analysis and assignee resolution as MCP tools. Drop them into any agent framework, or call REST directly.
Browse MCP servers →Insights
What the numbers mean for a filing decision
Three patterns matter for anyone deciding where to file next in RAG pipeline mechanics: how concentrated the field really is, which classes are getting crowded, and where the receiving-office pattern points.
Concentration
19.4% / 31.7%
top 5 / top 10 share of 489 records
Leadership is real but not closed
The leader holds 26 records and the field's top 5 combined account for 19.4% of all 489 records in scope; the top 10 reach 31.7%. That leaves roughly two-thirds of the record set outside the ranked leaders — a long tail of single- and few-filing entrants still has room to establish position.
Based on the 100-company assignee ranking returned by the dataset.
Technology density
73.8%
of records touch G06F
Core pipeline mechanics are the crowded ground
G06F claims — covering the general digital-data-processing mechanics of chunking, indexing and retrieval — sit in nearly three-quarters of records. G06N (AI-model computing) is the clear second cluster at 32.3%. Filing a broad mechanics claim here means competing against dense prior art rather than opening new ground.
Class shares sum above 100% because records can carry multiple IPC codes.
Filing momentum
261 in 2025
peak filing year
The field is still accelerating, not maturing
Filings moved from zero in 2017 to a peak of 261 in 2025. High density in the core classes reflects occupied claim space, not a technology that has settled — the pipeline mechanics are still being actively re-architected around adaptive chunking, context-window-aware processing and reranking.
2026 count is partial due to publication lag.
Eureka AI Agent
Looking for what nobody has claimed yet?
Eureka can read the same corpus for gaps instead of for coverage: under-claimed branches adjacent to retrieval augmented generation pipelines, with the prior art for and against each one.
Source: Patsnap Eureka. Co-assignee relationships and derived observations. Derived from a Patsnap search on Retrieval Augmented Generation Pipelines covering 2015–2026, data cut-off 2026-07-31. Counts reflect published records only and shift as new filings publish.Run this in Eureka MCP
Players
Who is filing, and what is still open
The ranked assignee list spans large technology platforms and smaller specialist filers. Recent-year momentum figures show single-digit filing counts even among ranked names, which is consistent with a field where the leaders have not yet locked down the pipeline's newer mechanics.
Leader
26 records
leader's record count
A single-digit-percentage leader
The top-ranked assignee holds 26 of the 489 records in scope — a meaningful lead over the field but far short of dominance. Recent-year momentum for several ranked names shows a drop to zero or one filing, suggesting this year's activity is spreading rather than consolidating.
Momentum figures reflect the latest partial filing year.
Mid-field
9 records
tenth-place record count
A shallow drop-off after the top ten
Tenth place holds 9 records, only a modest step down from fifth place at 16. The gap between the leader and the rest of the ranked group is wide, but the ranked group itself is fairly evenly spread — no single runner-up is close to catching the leader.
Based on the 100-company ranking returned by the dataset.
Receiving offices
319 US filings
largest receiving office
US-centric with a growing PCT track
United States filings account for the largest share of records at 319, with WIPO/PCT filings at 94 as the leading international route. India (29), Europe (22), Canada (9) and Australia (7) round out the filing geography, pointing to a field still primarily contested in the US before broader international filing.
Counts are by receiving office, not by assignee headquarters.
Recent-year filing momentum by assignee
| Assignee | Recent year | YoY |
|---|---|---|
| ISTARI DIGITAL INC | 1 | -50% |
| X Development LLC | 0 | -100% |
| Dell Products LP | 0 | -100% |
| Microsoft Technology Licensing, LLC | 0 | -100% |
| Amazon Technologies, Inc. | 0 | -100% |
| QOMPLX Inc. | 0 | -100% |
| Google LLC | 0 | -100% |
| Oracle International Corporation | 0 | -100% |
Source: Patsnap Eureka. Assignee-level momentum. Derived from a Patsnap search on Retrieval Augmented Generation Pipelines covering 2015–2026, data cut-off 2026-07-31. Counts reflect published records only and shift as new filings publish.Run this in Eureka MCP
What's Next
Where to take this analysis
The dataset points to specific next steps depending on whether the goal is freedom-to-operate, whitespace filing, or competitive tracking.
Check freedom-to-operate against the top 10
With 31.7% of records held by the top 10 assignees, any pipeline mechanics claim touching core chunking or context-window handling should be checked against this group's granted claims before drafting.
Run a claim comparison in EurekaScope a whitespace filing in the thinner classes
G16B, G10L and G06V each sit under 6% of records — far lighter coverage than the G06F/G06N core, and a plausible place to draft a first claim in a domain-specific application of RAG pipeline mechanics.
Explore whitespace with EurekaTrack momentum, not just rank
Several ranked assignees show a drop to zero filings in the latest partial year; rank alone does not show who is currently active. Monitoring year-over-year momentum is a better signal of who to watch next.
Set up assignee monitoring in EurekaSource: Patsnap Eureka. Forward-looking reading of the same dataset. Derived from a Patsnap search on Retrieval Augmented Generation Pipelines covering 2015–2026, data cut-off 2026-07-31. Counts reflect published records only and shift as new filings publish.Run this in Eureka MCP
FAQ
Common questions on RAG pipeline patents
This landscape identifies 489 published records matching claim language on RAG pipeline mechanics such as document chunking, chunk overlap, prompt construction, context window handling, relevance ranking and citation attribution, across coverage from 2015 through mid-2026. That figure reflects the specific search criteria used, not every AI-related filing that mentions retrieval — records without both a pipeline term and a technical control term are excluded. Because publication lags filing by roughly 18 months, the true number of filings made in 2025 and 2026 is understated in any current count.
The ranked assignee list contains 100 companies, with the leader holding 26 of the 489 records in scope. The top 5 assignees combined account for 19.4% of all records and the top 10 combined reach 31.7%, which is meaningful concentration but leaves roughly two-thirds of the field outside the ranked leaders. Recent-year momentum data shows even several ranked assignees dropping to zero or one filing, indicating the leadership picture is still moving rather than settled.
The dominant IPC class is G06F (electric digital data processing), appearing in 73.8% of the 489 records, followed by G06N (AI-model computing) at 32.3%. Smaller but distinct clusters appear in G06Q (business/admin processing), H04L (digital transmission), G10L (speech/audio), G16H (healthcare informatics), G16B (bioinformatics) and G06V (image/video recognition), each under 8.2% of records. Because a single record can carry multiple IPC codes, these shares add up to more than 100% of the record total.
Filing activity rose from effectively zero records in 2017 to a peak of 261 in 2025, indicating the field expanded sharply over that window. The 2026 count of 35 looks like a decline but is a partial-year figure caused by publication lag rather than a genuine drop in filing. A reliable growth rate cannot be computed from this data because fewer than four complete years remain once that lag is accounted for.
The lighter IPC clusters — G16B bioinformatics, G10L speech/audio and G06V image recognition, each under 6% of the 489 records — carry far less claim density than the core G06F and G06N classes. Sub-areas such as citation attribution scoring, index freshness and incremental re-embedding, and reranking-stage latency optimisation also show comparatively thin coverage relative to how central they are to a working RAG pipeline. These are reasonable starting points for a first claim, though any filing there should still be checked against the broader mechanics claims held by the top-ranked assignees.
Answers are grounded in the same dataset. Derived from a Patsnap search on Retrieval Augmented Generation Pipelines covering 2015–2026, data cut-off 2026-07-31. Counts reflect published records only and shift as new filings publish.Run this in Eureka MCP
Research Retrieval Augmented Generation Pipelines in depth with Eureka
Go past this page: query the whole retrieval augmented generation pipelines corpus yourself, in your own scope.
Every answer comes back with patent numbers you can open.
Disclaimer. This page is generated from Patsnap Eureka data drawn from a limited snapshot of global patent and scientific-literature records, and is provided for general information and reference only.
Patent data carries inherent limitations: recent filings (typically the most recent 18–24 months) are under-counted due to standard publication lag; counts may be reported at either a patent-family or a patent-record basis and are not always directly comparable; classification, applicant-name, and citation data may contain errors, duplicates, or omissions; and the underlying search query defines and constrains the scope shown. As a result, the analysis may be incomplete or inaccurate and may not reflect the full technology landscape.
Nothing on this page constitutes an exhaustive prior-art, novelty, freedom-to-operate, or validity search, nor does it constitute legal, financial, investment, or professional advice, and it should not be relied upon as such. Any patent, commercial, or strategic decision should be verified independently and reviewed with qualified patent, legal, and domain professionals. Patsnap makes no warranties, express or implied, as to the accuracy, completeness, or fitness for any particular purpose of the information presented.
Machine translation. Assignee and organisation names originally recorded in Chinese, Japanese or Korean have been rendered into English by an AI translation step so that the tables stay readable. These renderings are best-effort and may not match a company's registered English name; the original name is what the underlying patent record carries, and it is what any Eureka query launched from this page uses.