{
  "id": 8220422,
  "title": "An Introduction to Wontopos, the Memory API I Work On",
  "url": "https://urgent.news/2026/09/18/an-introduction-to-wontopos-the-memory-api-i-work-on",
  "topic": "ai",
  "section": "AI",
  "published": "2026-09-18T09:30:14.000Z",
  "source": {
    "name": "Dev.to",
    "slug": "dev-to",
    "url": "https://dev.to/woochan/an-introduction-to-wontopos-the-memory-api-i-work-on-4o1e"
  },
  "original_language": "en",
  "account": "Wontopos is a company that builds WOS, a long-term memory system for AI agents. The system stores an end user's memories once and recalls only the relevant ones when queried. This allows for efficient use of LLM prompts. Wontopos offers its services to both teams and individuals. For teams, there is an SDK that allows developers to decide exactly when to store data and what to recall. For individuals, there is a finished tool called MCP that can remember (such as Claude Code, Claude Desktop, Cursor, VS Code, Windsurf, Gemini CLI) without any integration code needed. The system ensures that each query returns a small, bounded set of information, regardless of how much is stored. This prevents the input cost from growing with the size of the store. Furthermore, the system is designed so that a store filled up for a year does not cost more to read from than a new one.\n\nThe models used by Wontopos are Tablet, Scroll, Tablet 1, Tablet 2, Scroll 1, Scroll 1.2, and Book. Tablet is the original model that returns fewer tokens and has no LLM involved. Tablet 2 is the default model and returns about 1,000 tokens per query. It introduces features that Tablet 1 does not, such as the ability to know who said what, store images as memories, and verify recall by checking additional memories if the initial attempt does not yield the answer. Scroll 1 adds an LLM query-understanding layer over the stored memories and reformulates the query only, returning about 3,700 tokens per query. Scroll 1.2, the current version of Scroll, performs sentence-level recall and returns about 2,800 tokens per query. The Book model is currently in development and has a fundamentally different design that aims to never repeat the same mistake. It is built for the goal of accuracy, rather than the trade-off between context and accuracy.\n\nAll measurements published by Wontopos on public benchmarks with every run reported can be found at https://wontopos.com/research.",
  "summary": "My posts here so far have only been about the benchmark I'm working on. This time I want to introduce the company I work at, Wontopos. The company Wontopos builds WOS: long-term memory for AI agents. It stores an end user's memories once, then recalls only the relevant ones per query so you can feed them into an LLM prompt. It is not only for teams. If you are building a product, you use the SDK…",
  "key_points": [],
  "editors_take": null,
  "illustration": null,
  "coverage": {
    "outlets": 1,
    "also_reported_by": []
  },
  "ai_generated": true,
  "disclaimer": "Summaries, key points and the editor’s take are written by software from other outlets’ reporting and may contain errors — always check the linked original."
}