Papers
arxiv:2611.12068

MIRIX-2.0: Multi-Agent Memory System

Published on Nov 9
ยท Submitted by
Weizhi Zhang
on Nov 16
Authors:
,
,
,
,

Abstract

MIRIX-2.0 is a multi-agent memory system for LLM agents that achieves high accuracy and efficiency across QA and procedural benchmarks without fine-tuning.

Long-term memory lets LLM agents personalize, stay consistent, and reuse what worked before. Parametric memories need training, consolidation can lose details for good, and keeping every record leaves a growing store to search. We present MIRIX-2.0, a multi-agent memory system designed to be practical, accurate, and efficient at once. It is the most accurate of the compared systems on four of five QA benchmarks. It reaches 91.4% on LoCoMo and 72.6% on LongMemEval-S, where it leads the strongest baseline by 14.3 points, and with AutoDream disabled it gains 31.0 and 22.0 points over HyperMem on single-hop and multi-hop document QA. As procedural memory, it yields the highest mean success among the compared memory methods on ALFWorld and WebShop and reaches 49.2% on MetaClaw-Bench, against 44.0% for MetaClaw's skill library. Six typed memory agents keep complete records as the system of record, and a provenance hypergraph indexes episodic and semantic records with n-ary frames that cite their rows. Retrieval runs under explicit budgets, so the evidence a question inspects does not grow with the history. AutoDream merges duplicate anchors outside the write path without rewriting any record, and it distills task sessions into skills. With AutoDream disabled, MIRIX-2.0 answers faster than HyperMem on all five timed benchmarks. It needs no fine-tuning, runs with closed and open models, and its code is public.

Community

Paper submitter

MIRIX-2.0 is out ๐Ÿš€ ๐Ÿš€ ๐Ÿš€

one system for both factual and procedural memory. Complete records as the system of record, a provenance hypergraph citing answers back to raw rows, and AutoDream merging duplicates and distilling skills offline, outside the write path. No fine-tuning. Numbers:

User Factual memory: AutoDream with 91.4 on LoCoMo and 72.6 on LongMemEval-S, +14.3 over the strongest baseline. AutoDream alone adds +3.0 on LoCoMo without rewriting a single stored record.

Agent Procedural memory: AutoDream skills lift ALFWorld to 76.4% vs 69.4 without memory, best of the memory methods on WebShop with the fewest tokens, 49.2% vs 44.0 on MetaClaw-Bench, file completion 27.7% vs 17.1%.

Keeping complete records as the system of record and letting typed agents do the retrieval feels like the right split between remembering everything and finding the right thing. Leading LongMemEval-S by 14 points is striking. The document QA gains are reported with AutoDream disabled, so what does it add when it is on?

Hi authors, thanks for the work. A question on Table 7: with majority-of-3 judging on 300 LongMemEval-S questions, accuracies should be multiples of 1/300. The baselines and M1 are, but M0 58.8 and M2 72.6 are not (nearest 58.67/59.00 and 72.33/72.67); both fit k/900. Were they scored differently? Details in this GitHub issue: https://github.com/Mirix-AI/MIRIX/issues/156

This is an automated message from the Librarian Bot. I found the following papers similar to this paper.

The following papers were recommended by the Semantic Scholar API

Please give a thumbs up to this comment if you found it helpful!

If you want recommendations for any Paper on Hugging Face checkout this Space

You can directly ask Librarian Bot for paper recommendations by tagging it in a comment: @librarian-bot recommend

Sign up or log in to comment

Get this paper in your agent:

hf papers read 2611.12068
Don't have the latest CLI?
curl -LsSf https://hf.co/cli/install.sh | bash

Models citing this paper 0

No model linking this paper

Cite arxiv.org/abs/2611.12068 in a model README.md to link it from this page.

Datasets citing this paper 0

No dataset linking this paper

Cite arxiv.org/abs/2611.12068 in a dataset README.md to link it from this page.

Spaces citing this paper 0

No Space linking this paper

Cite arxiv.org/abs/2611.12068 in a Space README.md to link it from this page.

Collections including this paper 0

No Collection including this paper

Add this paper to a collection to link it from this page.