AI conversation context vs Memory: what survives a summary

Your chat history can stay saved while an AI model works with a shorter version of the conversation. In noHuman Team, working context is the material the model uses for the current exchange. Memory contains notes saved outside that conversation. A conversation summary makes room in working context while the saved records stay on your computer.

That distinction matters when you ask a noHuman to work on a long project, return after a break, or continue a delegated task after a restart.

Three places your information can live

A model does not reread every saved file on every message. Keep the current task clear and point the noHuman to the source file when exact details matter.

What happens when a noHuman compacts its conversation?

The AI software summarizes older conversation content to create room for more work. The summary keeps a shorter account of the conversation, rather than an exact copy of each earlier message.

After compaction, noHuman Team sends the current working notes and open-task records again with the next turn. That gives the noHuman its saved task status alongside the summary.

For example, a Developer may have reported that it reproduced a checkout bug and is now testing a fix. Its open-task record can carry that report and the milestones named by the lead. The Developer can continue from there without relying on the entire earlier chat being present.

A summary can omit details. Keep an exact acceptance criterion, a final decision or a file you must preserve in Memory or the workspace, and tell the noHuman where to find it.

Choose when a quiet conversation gets summarized

From noHuman Team 4.7, open the noHuman's Settings → Work limits and find Compact when idle.

You choose how many quiet minutes to wait and how large the conversation must be. The defaults are 10 minutes and 100,000 tokens. The runtime leaves smaller conversations alone and waits while work is running or queued. Turn the setting off if you prefer to choose the timing yourself.

Save your changes when you are ready. The dashboard asks to restart that noHuman to apply its settings.

In the noHuman's Profile panel, Working context shows the current usage. Make room lets you request compaction yourself.

Should I compact after every task?

It depends on how that noHuman works. Compact after a delegated task is a separate option, off by default.

It can help when a worker finishes distinct assignments, such as checking a set of pages and then starting an unrelated review. The runtime waits briefly after completion so an immediate follow-up has time to arrive, and it leaves small conversations alone.

If you plan to ask detailed follow-up questions about the same exchange, you may prefer to keep its current context until you finish. You can leave after-task compaction off and use the idle setting or Make room later.

Does this lower my AI bill?

A shorter conversation can reduce the amount of context processed in later steps. Compaction itself also uses the AI connection to create a summary. Your provider's caching, model and pricing determine the cost of both operations.

noHuman Team 4.7 also keeps stable instructions separate from changing working notes. It sends the changed notes with the next message, which can avoid disturbing reusable context. This is a way to reduce repeated material, not a fixed savings promise.

Use the task's usage information to compare similar work before and after a change. A smaller context is useful only if the noHuman still has what it needs to complete the task.

Does a restart erase unfinished assignments?

noHuman Team saves task records on disk. Recent open tasks, their last reports and recorded milestones return with each turn, including after a restart or summary. Older open work remains available through the noHuman's work tools.

Those records describe reported progress. You still review the resulting files, tests or answers when the outcome matters.

Read what changed in noHuman Team 4.7, explore AI agent Memory, or install noHuman Team.