Engineering notes.
How jsonbolt is built, measured, and kept fast. No launch fluff.
What should your AI agent do with a giant JSON file?
jq and ripgrep vs the jsonbolt CLI on a 1 GB file the agent has never seen. Six schema-blind tasks, token counts included: ~40% fewer tokens, ~31× faster, zero silent failures.
Read post →The Dadroit JSON Viewer alternative for huge files: JSONBolt vs Dadroit.
A side-by-side comparison on the same machine: ~2.5× faster on a 1 GB file, YAML export, a CLI, dark mode, and key/value search Dadroit doesn't have. Plus the places Dadroit is still the better buy.
Read post →How we parse JSON at 3 GB/s.
Branchless structural detection, SIMD masks, lazy values, and the Bolt index. None of it is novel on its own; the win is in how the techniques compound when the parser is written for its consumer.
Read post →Feeding large JSON to an LLM without blowing the context window.
A 200 MB API response will eat your context window and your wallet. Token math, five compaction tricks, and a CLI pipeline that turns 487 MB into an 18k-token paste.
Read post →