Hugging Face Daily Papers · August 25, 2026 · 3 min read

Prime Agent: A Self-Improving RLM Harness

#agents #reasoning #open-source #security

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

Like Read original ↗

<a href=\"https://cdn-uploads.huggingface.co/production/uploads/6658e1c8ce1b2838885b2d7f/HDWYbg85SlEOUilNechaE.png\" rel=\"nofollow\"><img src=\"https://cdn-uploads.huggingface.co/production/uploads/6658e1c8ce1b2838885b2d7f/HDWYbg85SlEOUilNechaE.png\" alt=\"image\"></a></p>\n","updatedAt":"2026-08-25T03:08:16.773Z","author":{"_id":"6658e1c8ce1b2838885b2d7f","avatarUrl":"/avatars/8623555f14b62f40fd372da20cb59ccc.svg","fullname":"Seth Karten","name":"milkkarten","type":"user","isPro":false,"isHf":false,"isHfAdmin":false,"isMod":false,"followerCount":3,"isUserFollowing":false}},"numEdits":0,"identifiedLanguage":{"language":"en","probability":0.4194090962409973},"editors":["milkkarten"],"editorAvatarUrls":["/avatars/8623555f14b62f40fd372da20cb59ccc.svg"],"reactions":[],"isReport":false}}],"primaryEmailConfirmed":false,"paper":{"id":"2608.23552","authors":[{"_id":"6a8d069e5add2537c32e96d8","user":{"_id":"6658e1c8ce1b2838885b2d7f","avatarUrl":"/avatars/8623555f14b62f40fd372da20cb59ccc.svg","isPro":false,"fullname":"Seth Karten","user":"milkkarten","type":"user","name":"milkkarten"},"name":"Seth Karten","status":"claimed_verified","statusLastChangedAt":"2026-08-25T08:13:58.652Z","hidden":false},{"_id":"6a8d069e5add2537c32e96d9","name":"Alex L. Zhang","hidden":false},{"_id":"6a8d069e5add2537c32e96da","name":"Kevin Thomas","hidden":false},{"_id":"6a8d069e5add2537c32e96db","user":{"_id":"64f5e17e095cabd5e82e6fb5","avatarUrl":"/avatars/f82337208e4ea3e50394752a50c41e7e.svg","isPro":false,"fullname":"Sebastian Müller","user":"snimu","type":"user","name":"snimu"},"name":"Sebastian Müller","status":"claimed_verified","statusLastChangedAt":"2026-08-25T08:45:04.063Z","hidden":false},{"_id":"6a8d069e5add2537c32e96dc","name":"Elie Bakouch","hidden":false},{"_id":"6a8d069e5add2537c32e96dd","name":"Daniel Auras","hidden":false},{"_id":"6a8d069e5add2537c32e96de","name":"Mika Senghaas","hidden":false},{"_id":"6a8d069e5add2537c32e96df","name":"Fares Obeid","hidden":false},{"_id":"6a8d069e5add2537c32e96e0","name":"Konstantin Dunas","hidden":false},{"_id":"6a8d069e5add2537c32e96e1","name":"Johannes Hagemann","hidden":false},{"_id":"6a8d069e5add2537c32e96e2","name":"Sami Jaghouar","hidden":false}],"publishedAt":"2026-08-24T00:00:00.000Z","submittedOnDailyAt":"2026-08-25T00:00:00.000Z","title":"Prime Agent: A Self-Improving RLM Harness","submittedOnDailyBy":{"_id":"6658e1c8ce1b2838885b2d7f","avatarUrl":"/avatars/8623555f14b62f40fd372da20cb59ccc.svg","isPro":false,"fullname":"Seth Karten","user":"milkkarten","type":"user","name":"milkkarten"},"summary":"Language models are sequential processors, but long-horizon agency requires external information and computation beyond model weights and active context. Prime Agent is an open-source harness for long-horizon evaluation and coding-agent workflows. A persistent IPython REPL follows the Recursive Language Model abstraction for programmatic context processing and test-time compute, while Continual Harness preserves histories, memories, skills, prompts, and subagent specifications across trajectories. Recursive subagents coordinate through direct agent-to-agent communication, and the Agents View lets humans inspect and manage daemon-backed sessions. Prime Agent standardizes execution, recovery, verification, and resource accounting while leaving strategy construction to the model. This low-friction, expressive membrane prevents harness failures from becoming model failures and pushes measurement toward the model's true maximal underlying capability. Prime Agent raises ARC-AGI-3 RHAE Best@1 from 30% to 95.5% and matches or exceeds native and popular harnesses across long-context coding, GPU-kernel generation, emulator construction, and autonomous nanoGPT speedruns. On Factorio, we find refinement allows for continuous technology progression and dedicated subagents enable parallelized work. Code is available at https://github.com/PrimeIntellect-ai/prime-agent.","upvotes":18,"discussionId":"6a8d069e5add2537c32e96e3","projectPage":"https://www.primeintellect.ai/blog/prime-agent","githubRepo":"https://github.com/PrimeIntellect-ai/prime-agent","githubRepoAddedBy":"user","ai_summary":"Prime Agent is an open-source harness that uses recursive subagents, persistent computation, and agent-to-agent coordination to extend language models' long-horizon capabilities across coding and reasoning tasks.","ai_keywords":["Recursive Language Model","test-time compute","subagents","agent-to-agent communication","daemon-backed sessions","ARC-AGI-3","long-context coding","GPU-kernel generation","nanoGPT"],"ai_summary_model":"thinkingmachines/Inkling-Small","githubStars":18181,"organization":{"_id":"656ec1d908bd4deb79a0ba70","name":"PrimeIntellect","fullname":"Prime Intellect","avatar":"https://cdn-avatars.huggingface.co/v1/production/uploads/61e020e4a343274bb132e138/H2mcdPRWtl4iKLd-OYYBc.jpeg"}},"canReadDatabase":false,"canManagePapers":false,"canSubmit":false,"hasHfLevelAccess":false,"upvoted":false,"upvoters":[{"_id":"6658e1c8ce1b2838885b2d7f","avatarUrl":"/avatars/8623555f14b62f40fd372da20cb59ccc.svg","isPro":false,"fullname":"Seth Karten","user":"milkkarten","type":"user"},{"_id":"642c54a5b09c70b36de03071","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/642c54a5b09c70b36de03071/EwQyQmust01dgkdRMtYFt.jpeg","isPro":true,"fullname":"rasdani","user":"rasdani","type":"user"},{"_id":"64944cf67853dd12c3bcaf80","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/64944cf67853dd12c3bcaf80/NW2I8UqdDZwJTkIEYEeWU.jpeg","isPro":false,"fullname":"Mika Senghaas","user":"mikasenghaas","type":"user"},{"_id":"63445b570f69ad8aa6198f01","avatarUrl":"/avatars/a473341f8491ecf374e6c547b4858e5f.svg","isPro":false,"fullname":"Brendan Rappazzo","user":"bhogan","type":"user"},{"_id":"62bc74f868b72f4178beab5b","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/1664563862621-62bc74f868b72f4178beab5b.png","isPro":false,"fullname":"Garrett Goon","user":"garrett361","type":"user"},{"_id":"663058cca2b3e992a442c739","avatarUrl":"/avatars/acc4b29c4609e32d69d6b399052379a3.svg","isPro":false,"fullname":"Chengshuai Shi","user":"shichengshuai98","type":"user"},{"_id":"6039478ab3ecf716b1a5fd4d","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/6039478ab3ecf716b1a5fd4d/_Thy4E7taiSYBLKxEKJbT.jpeg","isPro":true,"fullname":"taesiri","user":"taesiri","type":"user"},{"_id":"67bcaba1608ec2bdb922e8ec","avatarUrl":"/avatars/ed6a3495586b3aaadd298a3d2d6cda59.svg","isPro":false,"fullname":"Alex Zhang","user":"a1zhang","type":"user"},{"_id":"63ac5701c21e60a3e9b58aa7","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/63ac5701c21e60a3e9b58aa7/g6EX7diOpuA94R2ab-rZC.png","isPro":true,"fullname":"Dipankar Sarkar","user":"dipankarsarkar","type":"user"},{"_id":"6342796a0875f2c99cfd313b","avatarUrl":"/avatars/98575092404c4197b20c929a6499a015.svg","isPro":false,"fullname":"Yuseung \"Phillip\" Lee","user":"phillipinseoul","type":"user"},{"_id":"651f8133dbf879b8c58f5136","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/651f8133dbf879b8c58f5136/0L8Ecgi5Ietkm_DchJwE-.png","isPro":false,"fullname":"Zikai Zhou","user":"Klayand","type":"user"},{"_id":"620783f24e28382272337ba4","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/620783f24e28382272337ba4/zkUveQPNiDfYjgGhuFErj.jpeg","isPro":false,"fullname":"GuoLiangTang","user":"Tommy930","type":"user"}],"acceptLanguages":["en"],"dailyPaperRank":0,"organization":{"_id":"656ec1d908bd4deb79a0ba70","name":"PrimeIntellect","fullname":"Prime Intellect","avatar":"https://cdn-avatars.huggingface.co/v1/production/uploads/61e020e4a343274bb132e138/H2mcdPRWtl4iKLd-OYYBc.jpeg"},"markdownContentUrl":"https://huggingface.co/buckets/huggingchat/papers-content/resolve/2608/2608.23552.md","query":{}}">

Papers

arxiv:2608.23552

Prime Agent: A Self-Improving RLM Harness

Published on Aug 24

· Submitted by

Seth Karten on Aug 25

Prime Intellect

Upvote

Authors:

Seth Karten ,

Sebastian Müller ,

Abstract

Prime Agent is an open-source harness that uses recursive subagents, persistent computation, and agent-to-agent coordination to extend language models' long-horizon capabilities across coding and reasoning tasks.

Generated by thinkingmachines/Inkling-Small

Language models are sequential processors, but long-horizon agency requires external information and computation beyond model weights and active context. Prime Agent is an open-source harness for long-horizon evaluation and coding-agent workflows. A persistent IPython REPL follows the Recursive Language Model abstraction for programmatic context processing and test-time compute, while Continual Harness preserves histories, memories, skills, prompts, and subagent specifications across trajectories. Recursive subagents coordinate through direct agent-to-agent communication, and the Agents View lets humans inspect and manage daemon-backed sessions. Prime Agent standardizes execution, recovery, verification, and resource accounting while leaving strategy construction to the model. This low-friction, expressive membrane prevents harness failures from becoming model failures and pushes measurement toward the model's true maximal underlying capability. Prime Agent raises ARC-AGI-3 RHAE Best@1 from 30% to 95.5% and matches or exceeds native and popular harnesses across long-context coding, GPU-kernel generation, emulator construction, and autonomous nanoGPT speedruns. On Factorio, we find refinement allows for continuous technology progression and dedicated subagents enable parallelized work. Code is available at https://github.com/PrimeIntellect-ai/prime-agent.