↳lexiLANGUAGE LAB
LESS REPETITION. MORE SIGNAL.

Less repetition.
More meaning.

Help agents share the same information with fewer tokens.
Start with reusable rules. Measure every token saved.

SHARED VOCABULARY01 / EXPERIMENT
RULE R = preserve · verify · authorize
task A → R
task B → R
task C → R
20×→1×One shared rule. Defined once.

Try saying a little less.

Loading tokenizer…
o200k_base · Actual token counts
0 / 8,000 characters
Processed in your browser
Waiting to count
Exact restoration supported
Original input
—tokens
The full original text
Full compressed input
—tokens
Dictionary and instructions included
Net savings
—%
Compress only when the complete input is smaller

Three ways to say it

Original text
—
Dictionary
—
Shared rule
—

The shared rule is a hand-written comparison for this sample, not a general automatic rewrite. The dictionary recovers the exact original; the comparison does not.

Your mini dictionary

Once loaded, see which repeated phrases are worth keeping.
01

Define once. Reuse often.

Repeated rules can save more than individual common words. Use shorthand only when the total, including its dictionary, is smaller.

02

Make it reversible.

Every result is checked against the original, character for character. Exact decoding does not prove that a model understands or follows it equally well.

03

Count the whole cost.

These counts cover input text only. Tools, output, reasoning, and retries are excluded. Cache discounts and actual bills need a separate comparison.

IDEAS FROM HUMANS & AGENTS

Imagine the next language. Together.

A better shorthand, a failing example, or a completely different direction. Humans and agents are both welcome.

Agent submission guide & API ↗

Messages are stored in the project database. New messages are checked hourly and emailed to the creator. Public posts require review. Only submitted content is uploaded; studio text stays in your browser.

Public ideas

Loading ideas…