parsec

Where the others stand

Every other layer stops at tool output. None of them curates thinking.

Only one of them (RTK) has been graded on whether the agent still solves the task, and it lost. Their headline numbers are single-session or single-scenario token counts; none reports task outcomes. Every Parsec number below is measured on one fixed model with the official SWE-bench harness, or on real agent traces with the same needed label the curator trains on.

Capability matrix · Parsec vs the field

Every other layer rewrites, summarizes, or deletes what your agent reads.

Capabilities and SWE-bench Verified results, Parsec vs Caveman, Headroom, Woz and RTK
 ParsecCavemanHeadroomWozRTK
Compresses tool outputyesyesyesyesshell only
Compresses thinkingyes no no no no
Trims the tool roster every turnyes no no no no
Never alters readable messagesyes no no no no
Kept content byte-exact (no rewording)yes no no no no
Learned from real agent tracesyes no no no no
Measured on the wire, not estimatedyesprovider-rep.estimatedestimatedisolated
Graded on task solves (SWE-bench, /100)6258585554
Cuts output tokens−43%−30% *−2%−1%+8%
Cuts input tokens (cache-aware)−54%−19%+6%−35%+16%
Cost per solved task$1.45$2.05$3.66$2.33$3.07
Mean added latency per call6.2 s6.9 s8.0 s14.3 s9.3 s
Runs localyesyesyes noyes

* Caveman cuts output by rewriting the reply into caveman-speak, not by reducing reasoning. Token, cost, time and solve columns are the six-arm SWE-bench Verified run (same model, official grader). Competitor claims as published on their sites and READMEs, 2026-09-09.

Rows = input types · columns = layers

Everyone shrinks some inputs. Only Parsec drops the thinking and the roster — and never a word you can read.

Share of each input type cut, Parsec vs Headroom, RTK, Caveman and Woz
 ParsecHeadroomRTKCavemanWoz
File reads82.2%47% not reported not reported not reported
Command output68.1% not reported89% not reported not reported
Search results98.2%92% not reported not reported not reported
Web content67.9% not reported not reported not reported not reported
Context & plans84.9% not reported not reported not reported not reported
Thinking69.7%0000
Tool roster77%0000
Your messages000 not reported0
Their own headline78.4%87%89%33%35%
Same meter — input tokens cut, SWE-bench−54%+6%+16%−19%−35%

Two rows have exactly one nonzero cell — thinking and the tool roster — both Parsec. On the same-meter row Parsec is −54%; Headroom and RTK add tokens.

Each competitor cell is that vendor's own published number, computed its own way (RTK's 89% is command output only — Read, Grep and Glob bypass its hook). Parsec cells are the share of that input type the agent never reused. The last row is one meter: provider-reported input tokens, same tasks, same model.

SWE-bench Verified · cost per task · Parsec vs no compression

Cost per task, every case.

Parsec drops the tool output and thinking the agent never uses. Across 100 SWE-bench Verified tasks that cut cost 39%; 21 cost more.

Cost per SWE-bench Verified task, no compression vs Parsec
SWE-bench Verified caseNo compressionParsecChange
matplotlib-20488$3.65$0.5286%
matplotlib-20826$4.11$1.2270%
django-11555$1.95$0.7661%
sympy-22456$6.24$2.7456%
django-15037$1.36$1.92+41%
sympy-23413$3.06$4.79+57%
psf/requests-5414$0.31$0.61+96%
All 100$147.30$89.65-39%

Parsec solved 12 the baseline missed; the baseline solved 7 Parsec missed. Net +5 solves at 39% lower total cost. We show the red rows.

The curator drops what the agent never reuses

Whatever the workload, ~three-quarters is never reused

Parsec's curator model scores every chunk of tool output and thinking and drops what the agent never reuses. The share it can drop lands near three-quarters no matter the workload.

Context saved by workload
WorkloadTracesContext inSentSaved
Software engineering3,363131,581,59225,765,83780.4%
Writing & docs1,76782,695,13217,873,23878.4%
Data analysis1,43660,466,98017,037,93271.8%
Research / web50910,731,2432,649,11375.3%
Devops70637,706,70011,077,92970.6%
Other20,706615,429,868128,598,90579.1%
All workloads28,487938,611,515203,002,95478.4%

Measured on 28,487 real agent traces.