Obsidian Agent Kit cuts coding-agent token usage by up to 53.2% in benchmarks
Scared_Resort_8177 · reddit · 2026-10-22
UPHASH tested the MIT-licensed Obsidian Agent Kit with Codex, Claude Code, and Cursor: agents read only relevant Markdown note sections, reuse known context, and verify whole-file hashes when work resumes — no embeddings or vector DB needed.
Token savings vs direct reads (36 sessions, 144 turns, best of 3 workloads per tool):
- Codex: 53.2%
- Claude Code: 48.0%
- Cursor: 40.7%
Methods, reproduction steps, and all results — including workloads that increased usage — are published on GitHub.
More from coding & agent
- Node.js ships security fixes across 22.x/24.x/26.x, worst bug allows RCE — dotey · 2026-10-22
- Research Progress Graph: modeling authentic research trajectories, 35 agents found incomplete — ZhiyuChen4 · 2026-10-22
- Longshot Studio 0.15 adds lip sync and face refinement, free on your own ComfyUI — R34vspec · 2026-10-22
- Dev builds Scratch-style MCP that feeds coding agents one task at a time to stop plan drift — NoTokenDev · 2026-10-22
- Frontier models spark a comeback of quirky game-style personal homepages — HamelHusain · 2026-10-22
- FOSS stack with DeepSeek beats paid ChatGPT plan in 2-month agent reliability test — layer4down · 2026-10-22