Intermediate · Systems

Agentic Context Management (ACM)

Visual diagram · Math · (in preparation) · Worked example · 3 difficulty levels.

TL;DR. The discipline of deciding what an AI agent knows at each step — assembling, curating, compacting, and securing its working context.

Technical Definition

The discipline of deciding what an AI agent knows at each step — assembling, curating, compacting, and securing its working context.

How it works

Agentic Context Management (ACM) is the practice of treating an agent's context window as a managed resource rather than an ever-growing transcript. An ACM layer decides what to include (instructions, tool schemas, retrieved documents, memory, prior results), what to summarize or drop, and what must never enter the window at all. It spans four concerns: relevance (does this token help the next decision?), cost and latency (tokens are money and milliseconds), integrity (is this content trustworthy?), and continuity (what must survive across sessions). Mature ACM implementations version context, log what was injected for auditability, and enforce trust boundaries between system instructions and untrusted retrieved text.

Visual Explanation (cycle)

Gather → Rank & filter → Compact → Inject into window → Act → Observe → Persist to memory → Gather

Related Concepts

• Agentic Context — The full set of information an agent has available when it decides its next action — goals, state, tools, memory, and retrieved data.
• Context Management — The engineering practice of controlling what goes into a model's context window, in what order, and at what cost.
• Context Compaction — Compressing accumulated history into a smaller, higher-density representation so an agent can keep working within its token budget.
• Context Security — Protecting the integrity and confidentiality of what enters and leaves an agent's context window.