Saving Tokens (and Money) in Claude

Three habits that keep your Claude usage low while getting the most out of Coppermind. None of them change what Coppermind can do -- they just stop you from spending tokens you do not need to.

Why This Matters#

Coppermind runs inside Claude, so your Claude plan covers the cost of talking to it. The more tokens a conversation burns, the faster you hit your plan limit -- and, if you have left one setting on, the more you can be charged on top of your plan. These three habits keep that under control.

1. Start a New Chat for Each Task#

This is the biggest one. Coppermind does not need conversation continuity to remember things. Everything it knows lives in the client's client mind, not in the chat history. So a fresh chat starts with the same full memory as a long one.

Long chats, on the other hand, get expensive fast. Every message you send re-sends the entire conversation so far, so a chat that has been going for an hour costs far more per message than a brand-new one -- for the exact same answer.

The habit: finished prepping for the Acme call? Start a new chat before you move on to the next client. You lose nothing -- ask "What do we know about [next client]?" and Coppermind picks up right where it should.

2. Use Sonnet for Daily Work, Opus for Strategy#

Claude lets you pick which model answers. The two you will use:

• Sonnet -- fast, capable, and much cheaper. Use it for almost everything: meeting prep, storing memories, searching, drafting follow-ups, daily briefings.
• Opus -- the most powerful (and most expensive) model. Save it for high-level strategy work: quarterly planning, complex positioning calls, deep document synthesis where the extra reasoning is worth the cost.

Most Coppermind work is retrieval and drafting, which Sonnet handles beautifully. Running everything on Opus is like taking a sports car to the grocery store -- you will burn through your plan limit far faster with no real benefit on routine tasks.

The habit: leave Claude on Sonnet by default. Switch up to Opus only when you are doing genuine strategic thinking, then switch back.

3. Check Your "Extra Usage" Setting -- This Is Real Money#

This one is about actual charges, not just speed. Read it carefully.

Claude has a setting variously called "Extra usage" or overflow billing. When it is turned on, Claude does not stop you when you reach your monthly plan limit -- it keeps going and charges you per token for everything past the limit. Those charges can add up quietly, especially if you have long chats running (see tip #1).

If you would rather hit your plan limit and stop -- with no surprise charges -- you need to turn this off.

The habit:

1. Open Claude's account or billing settings (claude.ai > Settings, or the equivalent in Claude Desktop)
2. Find the "Extra usage" / overflow / pay-as-you-go option
3. Turn it off unless you specifically want to pay for usage beyond your plan

Only leave it on if you have made a deliberate decision to accept per-token charges past your plan limit. For most beta testers, off is the safer default -- it means your bill is exactly your plan, never more.

Coppermind never charges you for Claude usage and cannot see or change this setting -- it lives in your Claude account. We are flagging it because it is easy to leave on by accident and only notice when the bill arrives.

Quick Reference#

Do these three and your Claude usage stays predictable -- no surprises, no wasted tokens, same full Coppermind memory every time.