Hugging Face Daily Papers · June 23, 2026 · 5 min read

OpenRath: Session-Centered Runtime State for Agent Systems

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

Like Read original ↗

OpenRath proposes a PyTorch-like programming model for multi-agent, multi-session systems, centered on a first-class <code>Session</code> runtime value.\nInstead of reconstructing transcripts, tool traces, memory events, workspace placement, and branch provenance from external logs, OpenRath carries them through the same runtime object used by agents and workflows. This makes fork, merge, replay, and inspection explicit and composable.\nCurious to hear feedback on the <code>Session</code> abstraction, the evidence protocol, and what evaluations or case studies would make this line of work more convincing.\n","updatedAt":"2026-06-23T03:27:43.886Z","author":{"_id":"68f792b148e9f4a46eca68bf","avatarUrl":"/avatars/a6442a2777a4b7e78f01168b80847446.svg","fullname":"Fukang Wen","name":"smallkang2025","type":"user","isPro":false,"isHf":false,"isHfAdmin":false,"isMod":false,"followerCount":2,"isUserFollowing":false}},"numEdits":0,"identifiedLanguage":{"language":"en","probability":0.8801875710487366},"editors":["smallkang2025"],"editorAvatarUrls":["/avatars/a6442a2777a4b7e78f01168b80847446.svg"],"reactions":[],"isReport":false}},{"id":"6a3ab3f98f5693f9d8e29641","author":{"_id":"6a029a5ef403289a6b544734","avatarUrl":"/avatars/8443ba67f1d74e5cb45aafa0fd63d82f.svg","fullname":"Jack","name":"AzenJack","type":"user","isPro":false,"isHf":false,"isHfAdmin":false,"isMod":false,"isUserFollowing":false},"createdAt":"2026-06-23T16:27:37.000Z","type":"comment","data":{"edited":false,"hidden":false,"latest":{"raw":"a nice work about multi agent！","html":"a nice work about multi agent！\n","updatedAt":"2026-06-23T16:27:37.395Z","author":{"_id":"6a029a5ef403289a6b544734","avatarUrl":"/avatars/8443ba67f1d74e5cb45aafa0fd63d82f.svg","fullname":"Jack","name":"AzenJack","type":"user","isPro":false,"isHf":false,"isHfAdmin":false,"isMod":false,"isUserFollowing":false}},"numEdits":0,"identifiedLanguage":{"language":"en","probability":0.7651076912879944},"editors":["AzenJack"],"editorAvatarUrls":["/avatars/8443ba67f1d74e5cb45aafa0fd63d82f.svg"],"reactions":[],"isReport":false},"replies":[{"id":"6a3abf7f2dfefd7b1648719b","author":{"_id":"68f792b148e9f4a46eca68bf","avatarUrl":"/avatars/a6442a2777a4b7e78f01168b80847446.svg","fullname":"Fukang Wen","name":"smallkang2025","type":"user","isPro":false,"isHf":false,"isHfAdmin":false,"isMod":false,"followerCount":2,"isUserFollowing":false},"createdAt":"2026-06-23T17:16:47.000Z","type":"comment","data":{"edited":false,"hidden":false,"latest":{"raw":"Thank you for your reply and for appreciating our work! 😄","html":"Thank you for your reply and for appreciating our work! 😄\n","updatedAt":"2026-06-23T17:16:47.352Z","author":{"_id":"68f792b148e9f4a46eca68bf","avatarUrl":"/avatars/a6442a2777a4b7e78f01168b80847446.svg","fullname":"Fukang Wen","name":"smallkang2025","type":"user","isPro":false,"isHf":false,"isHfAdmin":false,"isMod":false,"followerCount":2,"isUserFollowing":false}},"numEdits":0,"identifiedLanguage":{"language":"en","probability":0.9384582042694092},"editors":["smallkang2025"],"editorAvatarUrls":["/avatars/a6442a2777a4b7e78f01168b80847446.svg"],"reactions":[],"isReport":false,"parentCommentId":"6a3ab3f98f5693f9d8e29641"}}]}],"primaryEmailConfirmed":false,"paper":{"id":"2606.19409","authors":[{"_id":"6a362251db23715e9da12e25","user":{"_id":"68f792b148e9f4a46eca68bf","avatarUrl":"/avatars/a6442a2777a4b7e78f01168b80847446.svg","isPro":false,"fullname":"Fukang Wen","user":"smallkang2025","type":"user","name":"smallkang2025"},"name":"Fukang Wen","status":"claimed_verified","statusLastChangedAt":"2026-06-22T16:14:27.169Z","hidden":false},{"_id":"6a362251db23715e9da12e26","name":"Zhijie Wang","hidden":false},{"_id":"6a362251db23715e9da12e27","name":"Ruilin Xu","hidden":false}],"publishedAt":"2026-06-17T00:00:00.000Z","submittedOnDailyAt":"2026-06-23T00:00:00.000Z","title":"OpenRath: Session-Centered Runtime State for Agent Systems","submittedOnDailyBy":{"_id":"68f792b148e9f4a46eca68bf","avatarUrl":"/avatars/a6442a2777a4b7e78f01168b80847446.svg","isPro":false,"fullname":"Fukang Wen","user":"smallkang2025","type":"user","name":"smallkang2025"},"summary":"Modern agent systems often suffer from fragmented runtime state: transcripts, tool effects, memory events, workspace placement, branch provenance, and replay evidence are recorded separately and become difficult to inspect or reproduce. OpenRath addresses this issue with a PyTorch-like programming model for multi-agent, multi-session systems. The analogy concerns the role of a central first-class runtime abstraction, not tensor computation. Its core abstraction is Session, the runtime value passed between agents and workflows. A Session is branchable, inspectable, replayable, backend-aware, and composable. It records conversation chunks, sandbox placement, lineage metadata, token usage, pending work, and tool evidence, while defining where memory interactions enter the runtime record. Since this state is carried by the same value used in program execution, fork, merge, and replay become explicit runtime operations rather than states reconstructed from external traces. OpenRath further defines Sandbox, Tool, Agent, Memory, Workflow, and Selector, with Selector turning control flow into runtime-routed decisions. This report presents the programming model, architecture, audited milestones, and evidence protocol. Its claims are limited to controlled runtime properties, while broad quantitative comparisons, live-provider quality, optional-backend availability, and memory quality are left for follow-on evaluation. The central thesis is that Session provides agent systems with a first-class runtime value for auditable composition.","upvotes":68,"discussionId":"6a362251db23715e9da12e28","projectPage":"https://docs.openrath.com","githubRepo":"https://github.com/Rath-Team/OpenRath","githubRepoAddedBy":"user","ai_summary":"OpenRath introduces a PyTorch-like programming model for multi-agent systems using Session as a central runtime abstraction that enables explicit fork, merge, and replay operations while recording comprehensive execution state.","ai_keywords":["Session","Sandbox","Tool","Agent","Memory","Workflow","Selector","runtime abstraction","fork","merge","replay","auditable composition","execution state"],"ai_summary_model":"Qwen/Qwen2.5-Coder-32B-Instruct","githubStars":672},"canReadDatabase":false,"canManagePapers":false,"canSubmit":false,"hasHfLevelAccess":false,"upvoted":false,"upvoters":[{"_id":"68f792b148e9f4a46eca68bf","avatarUrl":"/avatars/a6442a2777a4b7e78f01168b80847446.svg","isPro":false,"fullname":"Fukang Wen","user":"smallkang2025","type":"user"},{"_id":"6a029a5ef403289a6b544734","avatarUrl":"/avatars/8443ba67f1d74e5cb45aafa0fd63d82f.svg","isPro":false,"fullname":"Jack","user":"AzenJack","type":"user"},{"_id":"623461fccd8a0462e55b3666","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/1647600114080-noauth.jpeg","isPro":true,"fullname":"Guian Fang","user":"Enderfga","type":"user"},{"_id":"67e271966a80ad6ca9ace199","avatarUrl":"/avatars/77e73628204a0daa68f76a14313d62eb.svg","isPro":false,"fullname":"Ningyuan Liu","user":"tracebackkkk","type":"user"},{"_id":"648130567d65d9ac1734457e","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/noauth/vOUOX_ukpz1bCkz4k-cYF.jpeg","isPro":false,"fullname":"Lockinwize Lolite","user":"mzwing","type":"user"},{"_id":"663316c3d5677e3287dabfd9","avatarUrl":"/avatars/8e578fdd110b3383fd528d6adc034a16.svg","isPro":false,"fullname":"RuikangHu","user":"xy3666","type":"user"},{"_id":"6999828fa16cc5bd0360767a","avatarUrl":"/avatars/9d77d6c6b89f0d8f7703bc9e8962940c.svg","isPro":false,"fullname":"be","user":"torisk","type":"user"},{"_id":"66d12a79960d799f1a10af99","avatarUrl":"/avatars/91e81e982513f385220d77ee6878941e.svg","isPro":false,"fullname":"AidenZhao","user":"AidenZhaoy","type":"user"},{"_id":"620783f24e28382272337ba4","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/620783f24e28382272337ba4/zkUveQPNiDfYjgGhuFErj.jpeg","isPro":false,"fullname":"GuoLiangTang","user":"Tommy930","type":"user"},{"_id":"6a1124dc106d74fadc5eaf46","avatarUrl":"/avatars/4fe8e58f000e92083d442b00f6b5eaf9.svg","isPro":false,"fullname":"ShuoyuChen","user":"shuoyu622","type":"user"},{"_id":"6728357c33d949c9caf9dd26","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/6728357c33d949c9caf9dd26/XrNyp9fgF3mGuSTDhTOnb.jpeg","isPro":false,"fullname":"Tokisakix","user":"Tokisakix","type":"user"},{"_id":"6a2da6c8ca070ee12c6e396c","avatarUrl":"/avatars/0355287dcabaa67dbc7f0b10b87451f9.svg","isPro":false,"fullname":"Joe Mama","user":"JoeMama123123123","type":"user"}],"acceptLanguages":["en"],"dailyPaperRank":2,"markdownContentUrl":"https://huggingface.co/buckets/huggingchat/papers-content/resolve/2606/2606.19409.md","query":{}}">

Papers

arxiv:2606.19409

OpenRath: Session-Centered Runtime State for Agent Systems

Published on Jun 17

· Submitted by

Fukang Wen on Jun 23

#2 Paper of the day

Upvote

Authors:

Fukang Wen ,

Abstract

OpenRath introduces a PyTorch-like programming model for multi-agent systems using Session as a central runtime abstraction that enables explicit fork, merge, and replay operations while recording comprehensive execution state.

Generated by Qwen/Qwen2.5-Coder-32B-Instruct

Modern agent systems often suffer from fragmented runtime state: transcripts, tool effects, memory events, workspace placement, branch provenance, and replay evidence are recorded separately and become difficult to inspect or reproduce. OpenRath addresses this issue with a PyTorch-like programming model for multi-agent, multi-session systems. The analogy concerns the role of a central first-class runtime abstraction, not tensor computation. Its core abstraction is Session, the runtime value passed between agents and workflows. A Session is branchable, inspectable, replayable, backend-aware, and composable. It records conversation chunks, sandbox placement, lineage metadata, token usage, pending work, and tool evidence, while defining where memory interactions enter the runtime record. Since this state is carried by the same value used in program execution, fork, merge, and replay become explicit runtime operations rather than states reconstructed from external traces. OpenRath further defines Sandbox, Tool, Agent, Memory, Workflow, and Selector, with Selector turning control flow into runtime-routed decisions. This report presents the programming model, architecture, audited milestones, and evidence protocol. Its claims are limited to controlled runtime properties, while broad quantitative comparisons, live-provider quality, optional-backend availability, and memory quality are left for follow-on evaluation. The central thesis is that Session provides agent systems with a first-class runtime value for auditable composition.

View arXiv page View PDF Project page GitHub 672 Add to collection

Community

smallkang2025

Paper author Paper submitter about 22 hours ago

OpenRath proposes a PyTorch-like programming model for multi-agent, multi-session systems, centered on a first-class Session runtime value.

Instead of reconstructing transcripts, tool traces, memory events, workspace placement, and branch provenance from external logs, OpenRath carries them through the same runtime object used by agents and workflows. This makes fork, merge, replay, and inspection explicit and composable.

Curious to hear feedback on the Session abstraction, the evidence protocol, and what evaluations or case studies would make this line of work more convincing.