Hugging Face Daily Papers · · 4 min read

Agent Memory Distillation: Empowering Small LLM Agents with Hierarchical Teacher Memory

Mirrored from Hugging Face Daily Papers for archival readability. Support the source by reading on the original site.

We propose <strong>Agent Memory Distillation (AMD)</strong>, a framework that <strong>transfers structured knowledge from a large teacher agent to a small student agent</strong> through hierarchical memory. </p>\n<p>AMD constructs three complementary memory types from successful teacher trajectories: <strong>Workflow</strong> memory encodes task-level strategies, <strong>Subtask</strong> memory provides concrete behavioral examples at an intermediate granularity, and <strong>Function</strong> memory captures per-function calling conventions and common pitfalls.</p>\n","updatedAt":"2026-08-11T04:49:41.205Z","author":{"_id":"66ed7737f2f27a5dfd81ef09","avatarUrl":"/avatars/f45eea356e92ac7b3db23c2c92dec9fa.svg","fullname":"Kangsan Kim","name":"KangsanKim71","type":"user","isPro":false,"isHf":false,"isHfAdmin":false,"isMod":false,"followerCount":17,"isUserFollowing":false}},"numEdits":0,"identifiedLanguage":{"language":"en","probability":0.8355698585510254},"editors":["KangsanKim71"],"editorAvatarUrls":["/avatars/f45eea356e92ac7b3db23c2c92dec9fa.svg"],"reactions":[],"isReport":false}}],"primaryEmailConfirmed":false,"paper":{"id":"2608.07169","authors":[{"_id":"6a7aa4e8019ce76dc7b3aaf3","name":"Taeil Kim","hidden":false},{"_id":"6a7aa4e8019ce76dc7b3aaf4","name":"Kangsan Kim","hidden":false},{"_id":"6a7aa4e8019ce76dc7b3aaf5","name":"Sung Ju Hwang","hidden":false}],"publishedAt":"2026-08-07T00:00:00.000Z","submittedOnDailyAt":"2026-08-11T00:00:00.000Z","title":"Agent Memory Distillation: Empowering Small LLM Agents with Hierarchical Teacher Memory","submittedOnDailyBy":{"_id":"66ed7737f2f27a5dfd81ef09","avatarUrl":"/avatars/f45eea356e92ac7b3db23c2c92dec9fa.svg","isPro":false,"fullname":"Kangsan Kim","user":"KangsanKim71","type":"user","name":"KangsanKim71"},"summary":"Memory systems have shown promise for improving agent performance, but their potential remains largely unexplored for small language models, which struggle to generate sufficient successful trajectories on their own. We propose Agent Memory Distillation (AMD), a training-free framework that transfers structured knowledge from a large teacher agent to a small student agent through hierarchical memory. AMD constructs three complementary memory types from successful teacher trajectories: Workflow memory encodes task-level strategies, Subtask memory provides concrete behavioral examples at an intermediate granularity, and Function memory captures per-function calling conventions and common pitfalls. Workflow and Subtask memories are injected proactively at the start of each task, while Function memory is retrieved reactively upon tool-calling errors. We evaluate AMD on three tool-use benchmarks using four student models (4B-8B parameters) with GPT-5-mini as the teacher, achieving average accuracy gains of 27.2%p, 11.2%p, and 3.4%p on AppWorld, BFCL V3, and ToolSandbox, while consistently outperforming existing memory-based baselines. Further analysis shows that Subtask memory contributes the largest gains, teacher effectiveness depends on both teacher capability and student compatibility, and 4B-sized students benefit most from AMD.","upvotes":31,"discussionId":"6a7aa4e8019ce76dc7b3aaf6","projectPage":"https://agent-memory-distillation.github.io","githubRepo":"https://github.com/taeilkim2465/agentic_memory_distillation","githubRepoAddedBy":"user","ai_summary":"Agent Memory Distillation improves small language model tool-use performance by transferring structured hierarchical memory from a large teacher agent without additional training.","ai_keywords":["Agent Memory Distillation","hierarchical memory","Workflow memory","Subtask memory","Function memory","tool-calling errors","teacher-student distillation"],"ai_summary_model":"thinkingmachines/Inkling-Small","githubStars":2,"organization":{"_id":"6475760c33192631bad2bb38","name":"kaist-ai","fullname":"KAIST AI","avatar":"https://cdn-avatars.huggingface.co/v1/production/uploads/6469949654873f0043b09c22/aaZFiyXe1qR-Dmy_xq67m.png"}},"canReadDatabase":false,"canManagePapers":false,"canSubmit":false,"hasHfLevelAccess":false,"upvoted":false,"upvoters":[{"_id":"66ed7737f2f27a5dfd81ef09","avatarUrl":"/avatars/f45eea356e92ac7b3db23c2c92dec9fa.svg","isPro":false,"fullname":"Kangsan Kim","user":"KangsanKim71","type":"user"},{"_id":"67864e969ade3b15efd4044b","avatarUrl":"/avatars/3d3fdcc111515be5652f97f16e7d521d.svg","isPro":false,"fullname":"Chanuk Lee","user":"tally0818","type":"user"},{"_id":"65b7c0eb8237bf70e416355d","avatarUrl":"/avatars/191540c8fe16329969882f0403de612e.svg","isPro":false,"fullname":"Degwie","user":"Sung10230","type":"user"},{"_id":"64e5a1cd4c20016ec9020ec8","avatarUrl":"/avatars/d7ffe7fbbe39c0a013375357457c57b3.svg","isPro":false,"fullname":"Soyeong","user":"starsuzi","type":"user"},{"_id":"64cfa0b9749587dbe01d0079","avatarUrl":"/avatars/93ca0a1d9c5578d052c5af0d4d1a0252.svg","isPro":false,"fullname":"Yumin Choi","user":"YuminChoi","type":"user"},{"_id":"65f06c8356cb8a32b41baf83","avatarUrl":"/avatars/ac66cc63f3abade4a859f7bf9357682a.svg","isPro":false,"fullname":"Jiongdao Jin","user":"jiongdao","type":"user"},{"_id":"695b465b44ee441b116e6ee1","avatarUrl":"/avatars/79db7b60c47b743003729920fe94cfb6.svg","isPro":false,"fullname":"Minjae Kim","user":"kmj200392","type":"user"},{"_id":"65633c5e84a9fbe322f87d81","avatarUrl":"/avatars/7233a555b43c669847a950ce5697c92c.svg","isPro":false,"fullname":"DongkiKim","user":"DongkiKim","type":"user"},{"_id":"63d797de3130cadcaf814de9","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/no-auth/SMlVP7geAbd-5I_CdTwT7.png","isPro":false,"fullname":"Myeongkyun Cho","user":"hestu","type":"user"},{"_id":"64b74920fe6a108d03fed767","avatarUrl":"/avatars/a2c05b809c36fa5fab8e1a43b3e67051.svg","isPro":false,"fullname":"Minki Kang","user":"Nardien","type":"user"},{"_id":"695c7a71cc6bc15f0eea3dee","avatarUrl":"/avatars/9ba8d7e4a9beb252333b448266f261b9.svg","isPro":false,"fullname":"Chaewoon Ki","user":"kichaewoon","type":"user"},{"_id":"65e5bd4568234ef5d6decadc","avatarUrl":"/avatars/c41095a946c0176b949c0b3566136c05.svg","isPro":false,"fullname":"Jaehyeong Jo","user":"harryjo97","type":"user"}],"acceptLanguages":["en"],"dailyPaperRank":0,"organization":{"_id":"6475760c33192631bad2bb38","name":"kaist-ai","fullname":"KAIST AI","avatar":"https://cdn-avatars.huggingface.co/v1/production/uploads/6469949654873f0043b09c22/aaZFiyXe1qR-Dmy_xq67m.png"},"markdownContentUrl":"https://huggingface.co/buckets/huggingchat/papers-content/resolve/2608/2608.07169.md","query":{}}">
Papers
arxiv:2608.07169

Agent Memory Distillation: Empowering Small LLM Agents with Hierarchical Teacher Memory

Published on Aug 7
· Submitted by
Kangsan Kim
on Aug 11
Authors:
,

Abstract

Agent Memory Distillation improves small language model tool-use performance by transferring structured hierarchical memory from a large teacher agent without additional training.

Memory systems have shown promise for improving agent performance, but their potential remains largely unexplored for small language models, which struggle to generate sufficient successful trajectories on their own. We propose Agent Memory Distillation (AMD), a training-free framework that transfers structured knowledge from a large teacher agent to a small student agent through hierarchical memory. AMD constructs three complementary memory types from successful teacher trajectories: Workflow memory encodes task-level strategies, Subtask memory provides concrete behavioral examples at an intermediate granularity, and Function memory captures per-function calling conventions and common pitfalls. Workflow and Subtask memories are injected proactively at the start of each task, while Function memory is retrieved reactively upon tool-calling errors. We evaluate AMD on three tool-use benchmarks using four student models (4B-8B parameters) with GPT-5-mini as the teacher, achieving average accuracy gains of 27.2%p, 11.2%p, and 3.4%p on AppWorld, BFCL V3, and ToolSandbox, while consistently outperforming existing memory-based baselines. Further analysis shows that Subtask memory contributes the largest gains, teacher effectiveness depends on both teacher capability and student compatibility, and 4B-sized students benefit most from AMD.

Community

We propose Agent Memory Distillation (AMD), a framework that transfers structured knowledge from a large teacher agent to a small student agent through hierarchical memory.

AMD constructs three complementary memory types from successful teacher trajectories: Workflow memory encodes task-level strategies, Subtask memory provides concrete behavioral examples at an intermediate granularity, and Function memory captures per-function calling conventions and common pitfalls.

Upload images, audio, and videos by dragging in the text input, pasting, or clicking here.
Tap or paste here to upload images

· Sign up or log in to comment

Get this paper in your agent:

hf papers read 2608.07169
Don't have the latest CLI?
curl -LsSf https://hf.co/cli/install.sh | bash

Models citing this paper

No model linking this paper

Cite arxiv.org/abs/2608.07169 in a model README.md to link it from this page.

Datasets citing this paper

No dataset linking this paper

Cite arxiv.org/abs/2608.07169 in a dataset README.md to link it from this page.

Spaces citing this paper

No Space linking this paper

Cite arxiv.org/abs/2608.07169 in a Space README.md to link it from this page.

Collections including this paper

No Collection including this paper

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

Discussion (0)

Sign in to join the discussion. Free account, 30 seconds — email code or GitHub.

Sign in →

No comments yet. Sign in and be the first to say something.

More from Hugging Face Daily Papers