Scarwise

Your AI agents make a mistake once. Not twice.
No rules to write – it learns by itself.

AI agents call tools all day: they run commands, read files, call MCP servers. When a call fails, the agent finds a form that works — and forgets it by the next conversation. Scarwise remembers the mistakes made on your computer: when any of your agents is about to repeat one, the call is stopped before it runs and the agent is told what worked here.

Why

• No rules to write: Scarwise learns by itself from what failed and what then worked.
• No wasted calls, no wasted tokens: a mistake already made is stopped before it runs, and you see what it saved.
• One lesson protects every agent: a mistake one agent made with a tool is not repeated by the others.
• Tools that do not work are recognized: the agent is told the likely cause — a missing sign-in, a service that does not answer, a defect of the server — instead of retrying again and again.
• One step to install: npm install -g scarwise connects your agents; npm uninstall -g scarwise leaves nothing behind.
• Your computer, your data: what Scarwise learns stays on your computer; only cleaned general patterns go to the shared library (on by default, off with scarwise share off).

Works with

Documentation on GitHub · npm · PyPI · Discussions