Urgent.News

What's breaking now, across thousands of outlets.

AI

Agents don't need memory, they need documentation

Article URL: https://liao.gg/blog/agents-dont-need-memory Comments URL: https://news.ycombinator.com/item?id=49945933 Points: 211 # Comments: 113

Memory plugins attempt to help agents understand projects by providing isolated snippets and inserting them into a vector database. However, this approach fails because agents don’t need memory; they need documentation. Memory plugins suffer from numerous issues, such as trying to capture more information, indexing better, and retrieving smarter.

Their focus is on capturing and recalling past conversations, rather than how people typically handle knowledge: by writing things down. Documents are often the only source of information in a project, yet they are frequently overlooked. Agents should not be required to search through millions of tokens of past conversations; instead, they should have access to a structured workspace containing instructions, specs, decisions, research, and any other relevant information.

With Operator Memory, agents can consult a Markdown brain containing these resources, updating them as needed before and after working on a task. This system eliminates the need for vector databases, embeddings, summarizers, and other token-burning background daemons, focusing solely on plain Markdown documents that can be read, updated, committed, and shared.

Over a year of using this approach has proven its effectiveness, with the agent able to read, update, and share information seamlessly. For those interested in trying Operator Memory, it is available for free and open source at https://github.com/aerovato/operator-memory.

Written by urgent.news from Hacker News Best's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.

This story

This is one outlet's version. Read the fullest account.

Every outlet that reported this, and when →

Read the original at liao.gg →

More in AI

DeepSeek 4.1: 256-Expert MoE, MTP-4x and DualPipe 2.0

O mercado global de inteligência artificial acaba de ser sacudido por um terremoto arquitetural de proporções históricas.

  • DeepSeek 4.1 model released with 670B parameters
  • MTP-4x and DualPipe 2.0 technologies quadruple decoding speed
  • 256 specialized and 1 isolated expert eliminate representation loss

CLAUDE.md: what it is, what to put in it, and examples

CLAUDE.md is a markdown file of instructions that Claude Code reads at the start of every session. You write it, and Claude treats it as context for the project: build commands, conventions and rules…

  • Claude reads CLAUDE.md at start of each session
  • CLAUDE.md location varies by scope: policy, user, project
  • File should be concise, specific, and version-controlled

More from Saturday 3 October →