Reduzir resultados que consomem muitos tokens

Whatever your tool returns becomes part of the model's context, and every character spends tokens. Bloated output is real cost. 💸

A giant blob can fill the model's limited context window, pushing out earlier messages and leaving less room to actually reason.

Send back just the fields the task needs. A focused result helps the model far more than dumping an entire raw record.

Strip internal ids, timestamps, and metadata the model will never use. Pick only the relevant keys before you return them.

For long results, return one page with a cursor instead of thousands of rows. The model can ask for more only if it needs it.

Set a hard limit on result length and truncate beyond it, so one runaway query cannot flood the model with megabytes of text.

Sometimes a short summary beats raw data, especially for logs or documents where the model only needs the gist, not every line.

Pretty-printed, deeply nested JSON wastes tokens. A flat, compact shape carries the same facts with far fewer characters.

For truly large content, return a resource link the model can read on demand instead of inlining the whole payload every call.

Long stack traces eat tokens fast. Return a short, clear error message and keep the full trace in your server-side logs.

Concise results save budget and sharpen reasoning. A focused reply gives the model exactly what it needs and nothing extra.

Why trim a tool's output before returning it?

You returned only needed fields, paginated, capped size, summarized, and offloaded big data, so results stay token-efficient. 🎯