Agent Engineering

Short-Term Memory = Context Window

The context window is the model's workspace: everything the model can see must be placed on this desk. Scroll down and watch how the workspace gets filled up.

Context Window Usage 0% 0 / 131,072 Token

⚠️ Compression strategy about to trigger

🖥 Model's Workspace (Context Window = 128K Tokens)

10 steps total · watch the workspace fill up　

Takeaway

Takeaway　The context window is the model's only working memory. 128K Tokens sounds large, but every turn requires resending the full message history. 10 rounds of deep dialogue plus a long document can easily fill it up. Managing context means managing the memory lifespan of your product.