Long chats, context limits and starting fresh

If you've worked with your agent for hours in one chat and it suddenly seemed to forget the plan, you're not imagining it. Very long chats get worse. The fix is simple: chat is for working, files are for keeping. Save what matters to a file, then start a fresh chat.

Why long chats get worse

Every AI model can only see a certain amount of text at once, called its context window. As a chat grows, the oldest messages get summarised or fall out of view. Your agent isn't ignoring you; it genuinely can't see that part any more.

That leads to the familiar signs:

• Decisions you agreed on an hour ago start getting ignored.
• The plan gets blurry, especially if you changed direction a few times.
• Answers get slower and less sharp.
• You see an error about the context limit being exceeded.

A long chat is not permanent memory. It's just a long chat. For how lasting memory works, see how your agent's memory works.

Save your work before starting fresh

Before you leave a long chat, have your agent write down what matters.

Then start a New chat from the left menu and pick up from the note:

No need to paste everything again. The note lives on your agent's computer, so any new chat can read it.

Habits that keep chats sharp

• One job per chat. Start a new chat for each separate task rather than running everything in one thread.
• Save as you go. After each decision, ask your agent to update the project file.
• Keep briefs in files. Put long instructions, specs and reference material in files and point your agent to them, instead of pasting them into chat again and again.
• Watch for the signs. When answers start drifting, don't push on. Save, then start fresh.

Shorter chats also use less AI usage, because the agent re-reads less text with every reply. See AI usage, models and top-ups.

Your chats list

Your chats are listed under Chats on the left of your agent's page.

• OpenClaw agents show conversations from every place you talk to it, labelled by where they came from, like Main, Telegram or Discord. Use Refresh chats to update the list and Load older chats to see more.
• Hermes agents let you rename and delete chats from the list.
• Earlier conversations holds chats from before your agent moved to its new hosting in October 2026. They're read-only. Click Start a new chat to carry on.

Chats in the dashboard and chats in Telegram or Discord are separate conversations. Something you said in Telegram isn't automatically in your dashboard chat, unless your agent saved it to memory.

Starting fresh in Telegram or Discord

In a messaging app, send /new to start a fresh conversation with your agent. If it doesn't answer straight away after /new, wait a few seconds and send your message again. If it still won't reply, see agent not responding or showing an error.

Frequently asked questions

What does "context limit exceeded" mean?

The chat has grown past what the AI model can see at once. Ask your agent to save a handover note, then start a New chat.

Will I lose my work if I start a new chat?

Not if it's saved to files. Anything only said in the old chat won't carry over, so ask your agent to save it first.

How long can a chat be before it gets worse?

It depends on the model and on how much text and how many files are involved. When your agent starts forgetting earlier decisions, it's time to save and start fresh.

Can I delete old chats?

On Hermes agents, yes, from the Chats list. OpenClaw chats can't be deleted from the dashboard yet.

Why can't I reply to my old chats?

Chats under Earlier conversations are from before the move to new hosting and are read-only. Click Start a new chat to continue.