Mi-Memory

A Lifecycle Memory Framework for Personal AI

Homepage • Overview • Architecture • Highlights • Results • Citation

---

📢 News

• 2026-08 — Added a standalone MemFuse/ project page for the MemFuse paper.
• 2026-07 — Technical report posted on arXiv: Mi-Memory: A Lifecycle Memory Framework for Personal AI.
• 2026-07 — Repository currently hosts the project introduction page, report PDF, and overview figures.

🧭 Overview

Project homepage: https://darwin-agent.github.io/Mi-Memory/

This repository currently hosts the Mi-Memory technical report, project homepage, and visual assets.

Personal AI is moving beyond chat-only interaction toward continuous services across phones, cars, homes, wearables, cameras, and tools. In this setting, memory cannot remain a cache of prior conversations: it must preserve durable user state, connect answers to multimodal and device evidence, support correction and forgetting, bound policy evolution, and remain deployable across edge/cloud constraints.

Mi-Memory is a lifecycle memory framework organized around four roles:

• Structure — memory runtime, storage hierarchy, retrieval, filtering, and context assembly.
• Expansion — multimodal and cross-device evidence acquisition.
• Evolution — diagnostic iteration, governed strategy updates, and rollback.
• Deployment — substrate-independent memory variants for constrained environments.

📐 Architecture

Mi-Memory links the lifecycle through a shared audit contract: typed evidence payloads preserve source identity and provenance, diagnostic traces localize evidence loss, strategy artifacts make memory-policy changes explicit, and gate/rollback records bound accepted evolution.

✨ Highlights

🧱 Composable memory structure	Multi-granularity storage with stage-level diagnostic traces.
🌐 Multi-source evidence	Dialogue, multimodal perception, cross-device events, and causal fusion.
🔄 Governed evolution	D²ACCI and E²MEND governed strategy evolution.
📦 Deployment flexibility	Full-stack cloud reference and repository-native lightweight memory variant.
📊 Comprehensive evaluation	Structure benchmarks plus module-level, transfer-feasibility, and design-level evidence.

📊 Results

The results below summarize evidence anchors from the technical report. They are reported role by role rather than as a single leaderboard, and evidence maturity differs across tracks. See the arXiv report for the authoritative numbers, protocols, evidence boundaries, and detailed analysis.

Benchmark	Track	Metric	Score
LoCoMo	Structure	Judge Accuracy	93.59%
PersonaMem-V2	Structure	Preference Accuracy	57.24%
LongMemEval	Structure	Judge Accuracy	87.47%
Mem-Gallery	Expansion	Judge Accuracy (3-vote)	89.15%
MemFuseBench	Expansion	Internal Fusion Score	35.2%
D²ACCI / E²MEND LoCoMo	Evolution	Offline Run Accuracy	94.74%
LiteMem / LoCoMo-aligned	Deployment	Transfer Score / Retention	90.81% / 90.0%

🔭 Outlook

Mi-Memory is an initial step toward auditable personal-AI memory infrastructure. Future work will focus on stronger causal attribution, propagation-complete forgetting, federated cross-device memory, scalable diagnostic evolution, and standardized memory contracts.

📁 Repository Contents

.
├── index.html             # GitHub Pages project homepage
├── paper.pdf              # Released technical report
├── figure/                # README-facing visual assets
├── LICENSE
└── README.md

This repository is an introduction page for the Mi-Memory technical report.

📖 Citation

If you find this work useful, please cite:

@techreport{mimemory2026,
  title       = {Mi-Memory: A Lifecycle Memory Framework for Personal AI},
  author      = {Darwin Agent Team},
  institution = {Xiaomi},
  year        = {2026},
  url         = {https://arxiv.org/abs/2607.18975}
}

License

This project is licensed under the MIT License.