14 Local Agents Switched from Markdown to TOON for Working Memory
imac · reddit · 2026-07-05
A local-first agent team switched the working memory of their 14 harnesses from Markdown to TOON (Token-Oriented Object Notation) in December 2025, publishing an honest review: based on 100 CRM records, TOON takes about 2068 tokens, saving 33% compared to row-object JSON and 58% compared to pretty-printed JSON. However, it only saves 4% compared to compact Markdown tables, and columnar JSON actually saves 4% more than TOON—an "unfavorable" conclusion the author disclosed transparently.
Under a 2500 token budget, TOON can hold all 100 rows and answer correctly, while row-object JSON gets truncated to 81 rows and confidently gives the wrong answer. The author also reviewed a TOON CLI issue deemed "not a bug" by the maintainer—which was actually their own dialect drifting from the spec. The machine had memorized the correct answer months prior, but it wasn't surfaced until a human editor reviewed it.
More from coding & agent
- 'My intelligence stack': developer splits AI coding work across Opus 5.5, Fable and Astra with human review gates — sh_reya · 2026-10-27
- Baseten webinar: evaluate open models for coding agents by benchmarking on your own merged PRs — baseten · 2026-10-27
- Thorsten Ball on working with agents: run more evals, automate everything, use the best models — omarsar0 · 2026-10-27
- DeepMind paper: LLMs can't sample randomly on their own — pass random numbers from code instead — rohanpaul_ai · 2026-10-27
- BELL: Open-Source Stack Combines Browserbase, Eve, iMessage and Stripe Link for Personal Agents — jeff_weinstein · 2026-10-27
- Legora runs 2,000 parallel agent evals on Modal Sandboxes for legal AI — blelbach · 2026-10-27