Skip to main content
WriterPatent grant

Writer was granted a patent for improving long-context AI retrieval.

Source: Autobound Signal API · 40 signals tracked for Writer · updated

Get Writer's live signals in Claude or ChatGPT, or by API

AskWhat changed at Writer this week? Use Autobound signals for writer.com.

What happened

In October 2026 Writer patented a method for processing lengthy inputs more effectively in large language model applications such as document analysis and search.

Source

patents.google.comOct 20, 2026

US patent US12769728

Better inference pattern for long context retrieval

Abstract

The present disclosure relates to a method and system for enhancing inference in large language models (LLMs) over long input sequences. A segmented inference strategy may be employed, wherein the long context can be divided and sequentially processed through a key-value (KV) cache of the LLM. At each step, the model may generate auxiliary outputs (or margins), which may include extractive summaries or intermediate signals based on the segment's relevance to an instruction. These margins may then be classified and selectively retained to guide final inference on the instruction. The retained margins may be prepended to the instruction to facilitate improved generation without modifying the model's internal weights. The disclosed approach provides efficient localization of relevant content, improves comprehension of extended contexts, and reduces computational overhead. Moreover, the disclosed technique is particularly effective for retrieval-based NLP tasks and supports long-context reasoning in LLMs while enhancing inference efficiency and user experience.

patents.google.com/patent/US12769728Read the full source

Extracted by Autobound

From the Signal API record
Signal
Patent grant

What this signalsA new patent often shows where a company is putting its R&D budget.

Grant date
Oct 20, 2026
Inventors
4
Tech area
software / computing
Patent number
US12769728

The full record

From the Signal API record

People

  • Melisa RussakInventor
  • Umar JamilInventor
  • Axel MagnusonInventor
  • Mateusz RussakInventor
  • Waseem AlShikhInventor
  • Kiran KambleInventor

Details

Primary CPC class
G06F 16/243 (Electric digital data processing)
USPTO assignee
Writer, Inc.
What the invention does
A technique that breaks long text into sections, keeps useful information in a running memory, and produces relevance cues to help the model retrieve the right details.

Topics and mentions

Tags

  • Intellectual Property
  • Research & Development
  • Artificial Intelligence
  • Machine Learning
  • Language Technology

Technologies named

  • large language models
  • long-context retrieval
  • KV caching
  • natural language processing

Extraction

Detected
Oct 21, 2026
signal_type
patents-company
signal_subtype
patentGrant

Use this data

Get every patent signal for Writer and the companies you sell to, in the tools you already use.

  1. Ask Claude about it

    Connect Autobound to Claude, Claude Code or Cursor with MCP. Then ask: “What changed at Writer this week?”

  2. Send it to your own tools

    The Signal API returns patent signals for any list of companies as JSON, for your CRM, warehouse or app.

  3. Try it free

    Sign up and spend your free credits on the companies you sell to.

    Start Free1,000 free credits

The API returns more than this page shows

This page shows a preview. The full patents-company record in the Signal API and MCP can also have these 8 fields. Some fields are empty for some signals.

Company

  • linkedin_urlValue in the API
  • industriesValue in the API
  • employee_count_lowValue in the API
  • employee_count_highValue in the API
  • revenueValue in the API
  • descriptionValue in the API

Signal

  • signal_nameValue in the API
  • associationValue in the API
Show the full JSONThe record on this page and the API request

GET /v1/signals/ba0abfff-28b8-567d-a41f-5186c812ccdc returns this record as JSON. POST /v1/companies/enrich returns every signal for writer.com.

{
  "signal_id": "ba0abfff-28b8-567d-a41f-5186c812ccdc",
  "signal_type": "patents-company",
  "signal_subtype": "patentGrant",
  "detected_at": "2026-10-21T15:53:26+00:00",
  "company": {
    "name": "Writer",
    "domain": "writer.com"
  },
  "data": {
    "cpc": [
      {
        "code": "G06F 16/243",
        "label": "Electric digital data processing"
      },
      {
        "code": "G06F 40/284",
        "label": "Electric digital data processing"
      },
      {
        "code": "G06F 40/40",
        "label": "Electric digital data processing"
      }
    ],
    "tags": [
      "Intellectual Property",
      "Research & Development",
      "Artificial Intelligence",
      "Machine Learning",
      "Language Technology"
    ],
    "detail": "In October 2026 Writer patented a method for processing lengthy inputs more effectively in large language model applications such as document analysis and search.",
    "summary": "Writer was granted a patent for improving long-context AI retrieval.",
    "event_at": "2026-10-20",
    "inventors": [
      {
        "name": "Melisa Russak",
        "status": "resolved",
        "is_primary": true,
        "linkedin_url": "https://www.linkedin.com/in/melisa-russak-5b7987145"
      },
      {
        "name": "Umar Jamil",
        "status": "resolved",
        "is_primary": false,
        "linkedin_url": "https://www.linkedin.com/in/ujamil"
      },
      {
        "name": "Axel Magnuson",
        "status": "resolved",
        "is_primary": false,
        "linkedin_url": "https://www.linkedin.com/in/axel-magnuson-16286028"
      },
      {
        "name": "Mateusz Russak",
        "status": "unresolved",
        "is_primary": false
      },
      {
        "name": "Waseem AlShikh",
        "status": "resolved",
        "is_primary": false,
        "linkedin_url": "https://www.linkedin.com/in/waseemalshikh"
      },
      {
        "name": "Kiran Kamble",
        "status": "resolved",
        "is_primary": false,
        "linkedin_url": "https://www.linkedin.com/in/kiran-kamble07"
      }
    ],
    "tech_area": "software / computing",
    "event_kind": "grant_date",
    "grant_date": "2026-10-20",
    "source_url": "https://patents.google.com/patent/US12769728",
    "cpc_primary": {
      "code": "G06F 16/243",
      "label": "Electric digital data processing"
    },
    "patent_title": "Better inference pattern for long context retrieval",
    "patent_number": "US12769728",
    "uspto_assignee": "Writer, Inc.",
    "patent_abstract": "The present disclosure relates to a method and system for enhancing inference in large language models (LLMs) over long input sequences. A segmented inference strategy may be employed, wherein the long context can be divided and sequentially processed through a key-value (KV) cache of the LLM. At each step, the model may generate auxiliary outputs (or margins), which may include extractive summaries or intermediate signals based on the segment's relevance to an instruction. These margins may then be classified and selectively retained to guide final inference on the instruction. The retained margins may be prepended to the instruction to facilitate improved generation without modifying the model's internal weights. The disclosed approach provides efficient localization of relevant content, improves comprehension of extended contexts, and reduces computational overhead. Moreover, the disclosed technique is particularly effective for retrieval-based NLP tasks and supports long-context reasoning in LLMs while enhancing inference efficiency and user experience.",
    "precision_class": "structural",
    "invention_explanation": "A technique that breaks long text into sections, keeps useful information in a running memory, and produces relevance cues to help the model retrieve the right details.",
    "technologies_mentioned": [
      "large language models",
      "long-context retrieval",
      "KV caching",
      "natural language processing"
    ]
  }
}

Long text fields are shortened on this page.

Looking up one signal by its id is free. Enrich costs 2 credits per signal returned; a call with no results is free.