> ## Documentation Index > Fetch the complete documentation index at: https://docs.daoco.org/llms.txt > Use this file to discover all available pages before exploring further. # Context management > How daoco manages long conversations with automatic context compaction and thread hygiene. Each chat thread has a context window: the amount of conversation daoco can actively consider. The **context meter** in the composer shows how full that window is. When the thread has billed usage, the same circle also shows how many basic tokens the thread has charged so far. Hover the meter to see **current context** pressure (percentage, progress bar, and tokens in context vs the limit) plus cumulative **`in`**, **`out`**, and **`total`** counts for the thread. See [Understand token usage](/account-and-billing/understand-token-usage#context-meter-in-chat) for a full breakdown. ## Automatic compaction When the context window fills, daoco summarizes older messages so the thread can continue. The summary preserves important decisions, brand corrections, active drafts, file references, and outcomes from recent specialist runs. You may briefly see a compaction status while this happens. After compaction, earlier detail is condensed, but daoco keeps the thread usable for follow-up work. ## Active runs and queued follow-ups If you send a follow-up while daoco is still working, it appears immediately as a queued message instead of starting a competing run. When the active run is safe to steer, a plain-text follow-up can guide its next step. Otherwise, the message stays queued and is picked up at a safe stopping point. Compatible plain-text follow-ups can be combined into one turn. Attachments and structured inputs stay separate. When you pause a run yourself, queued follow-ups wait until you resume. If you cancel a queued follow-up before it starts, it is not sent to the model. ## What stays in context daoco prioritizes recent messages, active drafts, open approvals, attached assets, and durable workflow state such as campaign or post panels. Long research transcripts and large tool results may be summarized or trimmed before older casual chat history. ## When to continue a thread * You are refining the same post, campaign, asset, or clip job. * Earlier decisions still matter. * You want daoco to remember constraints from the conversation. * A specialist run already produced evidence or drafts you want to build on. ## When to start a new thread * The task is unrelated to the current work. * You want different assumptions or tone. * The current thread contains old constraints that could confuse the next task. * You finished a major workflow and want a clean slate for a new launch or campaign. Shorter, focused threads are easier to steer and usually more token-efficient. See [Understand token usage](/account-and-billing/understand-token-usage).