#contextwindow

Chunking, map-reduce, retrieval and compaction: how agents handle work that never fits in one context window, and keep their own history lean.

从零手写 Agent（三）：上下文超长，别等模型替你截断 第二篇里，我做了一个比较极端的压测。 把 Ollama 的 num_ctx 调得很小，然后不断往 messages...

从零手写 Agent（二）：短期记忆，其实就是你一直带着的 messages 上一篇把 Tool 跑通以后，我开始给 Agent...

8 Benefits Nobody Talks About There's more to token optimization than a smaller API bill....

You've hired a goldfish to guide you through the aquarium. He's enthusiastic, knowledgeable, and...

"Context engineering" is getting used as a rebrand for prompt writing, and that undersells it into uselessness. The actual discipline is deciding, every single turn, what earns a place in a bounded window and what gets evicted — and living with the fact that every eviction is a bet you're making about the future of a run you haven't finished yet. Here's the physical constraint underneath the buzzword, the packing decision that's actually hard, what it costs to get it wrong, and how to test whether your packing policy works.

In this article, we review Context-Mode. You will learn: What is Context-Mode? The problem How...

The documented target is under 200 lines per CLAUDE.md file. What that number means, the separate 4 MiB limit, and whether to trim, split or move cont

A Claude Code statusline can do far more than echo a model name. Color the context-remaining number and you /compact on your own terms instead of the tool’s.

Context compaction quietly drops the rules you give a coding agent. Two papers put numbers on it (17% and 53% retention). Here's what I moved into CLA

A larger context window gives an AI model more information to work with. It doesn't give the model a...

What fits, what falls out, and why long conversations get weird.