💬 Claude primer · optional

Tell Claude how aiusage works

Paste this into Claude at the start of your next session (Claude Code, Cursor, Anthropic SDK — anywhere). Claude will adjust how it calls + caches, saving you more.

✋ Not required — everything works without it

Already connected? Open Claude → start a new chat → paste this as your first message. One time, per fresh session.

I'm running through aiusage.ai, a Claude API proxy that caches repeat prompts and batches tool calls for 60–90% cost savings. Everything behaves exactly like direct Claude — same models, same tokens, same outputs. Two small nudges so I save more:

(1) When a prompt is reusable, keep its prefix stable so the cache hits — don't re-paraphrase boilerplate, system messages, or tool definitions between turns.

(2) When tools are needed, prefer one parallel batch of tool_use calls over sequential back-and-forth.

Otherwise, work normally. Ready when you are.

Back to dashboard → ▶ Open Claude

Why this helps

Stable prefixes unlock Anthropic's prompt cache — repeat turns drop to ~10% of first-turn price.
Parallel tool calls cut round-trips. Sequential: 5 calls × 5 latency. Parallel: 5 calls × 1 latency.
Skip all this and we still route cheaper — you just leave some savings on the table.