TRACE: Open-Source Hierarchical Memory for LLM Agents
PsychologicalDot7749 · reddit · 2026-07-06
TRACE is an open-source hierarchical memory system for LLM agents (pypi package trace-memory) that organizes conversation history into a topic tree (branches + summaries) rather than flat RAG chunks. The author evaluated it on the EventQA precise retrieval task of MemoryAgentBench (ICLR 2026).
F1 Results: TRACE achieved 82.5% with gpt-oss-20B and 83.8% with gpt-oss-120B. In contrast, Mem0 (GPT-4o-mini, official paper figures) scored 37.5%, and MemGPT/Letta (GPT-4o-mini) scored 26.2%.
It should be noted that the author ran the open-weight gpt-oss models locally, which isn't a fair backbone comparison against the GPT-4o-mini used by Mem0/MemGPT. The author attempted to run Mem0 directly on gpt-oss-20B for a fair comparison, but its fact extraction step requires strict JSON output, which gpt-oss failed to parse cleanly (a similar issue occurred with Gemini/Mistral). Letta was skipped as it requires a full server deployment. Complete JSON logs from two runs are open-sourced in the repository.
Related event: TRACE Open-Sources Hierarchical Memory for Agents(3 posts)→
More from coding & agent
- Claude Code + Monid MCP builds a $12K/year B2B prospecting engine on live data — aryanXmahajan · 2026-10-26
- Karpathy shares CLAUDE.md rulebook to stop AI coding agents from piling up tech debt — mdancho84 · 2026-10-26
- Rethinking dependencies in the agent era: drop libs that save 20 lines of code — gethackteam · 2026-10-26
- RustGIS hits 2,000 GitHub stars in two weeks with agent-first architecture — giswqs · 2026-10-26
- Vercel Ships 12 New Products at Its 2026 Conference as It Races Rivals for the Agent Entry Point — FinanceYF5 · 2026-10-26
- Build an AI agent from scratch in under 50 lines of pure Python: ReAct loop tutorial — adnan_hashmi · 2026-10-26