PasteCB — Smart macOS Clipboard for Agents

PasteCB is a smart clipboard for macOS, optimized for working with agents. It cuts vision token usage by 29% to 59% on every image you copy.

Download: Latest build (.zip)

---

Evals

Model: gpt-4o

Eval    Variant   Pixels       Size  Prompt  Total  Saved    File
------  --------  ---------  ------  ------  -----  -------  -------------------------------
eval-0  original  1244x808   62.1 KB    1134   1135  -        img-eval-0-1244x808.png
eval-0  PasteCB   700x455   104.1 KB     454    455  +59.96%  img-eval-0-pastecb-700x455.png
eval-1  original  1538x324   69.7 KB     794    795  -        img-eval-1-1538x324.png
eval-1  PasteCB   1000x211   85.5 KB     454    455  +42.82%  img-eval-1-pastecb-1000x211.png
eval-2  original  2172x1236 167.8 KB    1134   1135  -        img-eval-2-2172x1236.png
eval-2  PasteCB   1000x569  126.0 KB     794    795  +29.98%  img-eval-2-pastecb-1000x569.png

Average prompt-token savings across the 3 evals: +44.25%.

---

Run the evals

Run pastecb_vision_token_eval.py to reproduce the evals locally.

Setup

cd evals
python3 -m venv .venv
source .venv/bin/activate
pip install -r requirements.txt
# edit .env and set OPENAI_API_KEY

Run

python pastecb_vision_token_eval.py

Environment

Variable	Required	Default	Notes
OPENAI_API_KEY	yes	—	Loaded from evals/.env.
OPENAI_VISION_MODEL	no	gpt-4o	Any vision-capable chat model. CLI wins.

The script uses max_tokens=8 so totals stay dominated by prompt / image tokenization. Exact vision math is provider-specific; the usage field returned per request is what gets billed.

---

© 2026 macosfy. All rights reserved.