Plain Markdown Wiki Beats All AI Memory Products Across 2176 Tasks
Major-Shirt-8227 · reddit · 2026-08-03
A developer benchmarked 8 AI agent memory systems across 2176 tasks. Surprisingly, a plain markdown wiki curated by the agent itself (following Karpathy's gist) scored 98.5, beating every specialized product.
Key Findings:
- Mitosis Cortex ranked first among hosted products at 96.9.
- gbrain, a free local tool, scored 92.9, beating all hosted APIs except Mitosis.
- Mem0 (92.3) was the cheapest per 1,000 successful answers at $341.
- Zep struggled with freshness, taking a median of 162.7 seconds for a newly stored fact to become queryable.
- Supermemory had near-perfect recent recall (59/60) but failed long-horizon questions (11/72).
Recommendations: Mitosis Cortex for hosted APIs; Markdown files or gbrain for free, local setups.
More from coding & agent
- AI Becomes a Lifesaver for Resolving Complex Merge Conflicts — haltakov · 2026-08-03
- Eve framework: batteries-included AI agent deployment with durable execution and sandbox — evilrabbit_ · 2026-08-03
- Decode: A Whiteboard for Reviewing and Guiding AI Coding Agents — Vjeux · 2026-08-03
- Prompting AI to Build a Zero-Asset Dungeon Crawler with Fluid Simulation — Dimillian · 2026-08-03
- OpenWiki 0.2.5 Introduces `.openwikiignore` to Save Tokens — LangChain · 2026-08-03
- AI Agent Architecture Evolution: Separating Tools from Brains Redefines the Computer — irvinebroque · 2026-08-03