terminal
Same answer. Far fewer tokens.
Miru returns tight, ranked context instead of dumping files into the prompt.
you: where is auth handled?
Without Miru0
Miru0
Measured as tokens returned to the agent per search.
0%
Use your IDE as usual.
Miru plugs into MCP. Your agent keeps the same habits.
terminal
- Cursor
- Claude Code
- Gemini CLI
- Copilot
- Windsurf
- Codex
- Kiro
- and more
Your code stays yours.
terminal
- $ miru/
- $ auth/
- $ billing/
- $ index.bin
Local
Miru runs on your machine.
Global
Built for agentic demand.
Our embedding API keeps up. We never log or keep queries or code.
Don't trust our numbers? Benchmark Miru on your own code and ask your agent.
1
Turn it on
Run this in your terminal after installing Miru. Adds --benchmark to your installed MCP configs.
miru benchmark on2
Restart your IDE/agent
So it picks up the new benchmark flag.
3
Use your agent as normal
Every search and locate call quietly logs its savings.
4
Then you can ask
Ask any time, in plain language.
"how many tokens have I saved?"terminal