TALARIA
Token Efficiency Protocol
▸100% FREE · NO SUBSCRIPTION▸DELTA READS ACTIVE▸~40% FEWER TOKENS ON RE-READS▸COMPACT GUTTER {i}|{line}▸ON-CHAIN VERIFIED▸OPEN SOURCE▸BASE MAINNET · CHAIN ID 8453▸17 FILE-TOOL TESTS GREEN▸IMMUTABLE AUDIT TRAIL▸100% FREE · NO SUBSCRIPTION▸DELTA READS ACTIVE▸~40% FEWER TOKENS ON RE-READS▸COMPACT GUTTER {i}|{line}▸ON-CHAIN VERIFIED▸OPEN SOURCE▸BASE MAINNET · CHAIN ID 8453▸17 FILE-TOOL TESTS GREEN▸IMMUTABLE AUDIT TRAIL
1 · TOKEN EFFICIENCY — LIVE STATS
Stop paying for tokens your agent never reads.
TALARIA patches tools/file_operations.py with two changes: Delta Reads (re-reads send only changed hunks, not the whole file) and a compact gutter format. Together they cut up to 40% of token spend on iterative editing loops — the most common thing an autonomous agent does.
0
Tokens Saved
session total
0
File Reads
delta-compressed
0.0%
Efficiency Gain
vs padded gutter
$0.00
Cost Saved
USD equivalent
GUTTER FORMAT COMPARISON
BEFORE — padded+48% vs bare
␣␣1␣␣␣import os ␣␣2␣␣␣def main(): ␣34␣␣␣␣␣␣return x
leading spaces tokenize into extra tokens on EVERY line
NOW — compact + delta+16% vs bare · OPTIMAL
1| import os 2| def main(): 34| return x
numbers stay (model references them) · padding goes
A/B RESULT TABLE — Sonnet 4.6, 2 passes, 4-task battery
VARIATIONRESULTNOTESΔ TOKENS
padded (legacy)4/4 PASSbaseline format+48%
compact4/4 PASS ✓numbers referenced correctly+16%
compact + delta4/4 PASS ✓re-reads send only diffs−24%
no numbers3/4 PASSoff-by-one, hand-counted0%
5 · UNAFFECTED BY FORMAT CHANGE
✓patch / fuzzy_match — match text, never consume gutter
✓no downstream parser keys on fixed-width columns
✓editing behavior identical
✓blockchain hash verification unaffected