https://github.com/exoharness/exo/
https://github.com/laude-institute/headlong
https://github.com/microsoft/agent-lightning
and now https://github.com/PrimeIntellect-ai/prime-agent
Of course the don't have exactly the same scopes but they are in general all about persistent memory and / or continous agent loops. Like I miss those times where only once a week a new js framework was promoted."},"story_title":{"matchLevel":"none","matchedWords":[],"value":"Headlong: A microharness for persistent agents"},"story_url":{"matchLevel":"none","matchedWords":[],"value":"https://www.laude.org/updates/headlong-a-microharness-for-persistent-agents"}},"_tags":["comment","author_ma2kx","story_49428882"],"author":"ma2kx","children":[49431639],"comment_text":"I'm just exhausted. So I've today now learned about four new harness:
https://github.com/exoharness/exo/
https://github.com/laude-institute/headlong
https://github.com/microsoft/agent-lightning
and now https://github.com/PrimeIntellect-ai/prime-agent
Of course the don't have exactly the same scopes but they are in general all about persistent memory and / or continous agent loops. Like I miss those times where only once a week a new js framework was promoted.","created_at":"2026-08-25T05:45:52Z","created_at_i":1787636752,"objectID":"49429553","parent_id":49429134,"story_id":49428882,"story_title":"Headlong: A microharness for persistent agents","story_url":"https://www.laude.org/updates/headlong-a-microharness-for-persistent-agents","updated_at":"2026-08-25T23:18:19Z"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"hedgehog"},"comment_text":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["prime","agent","primeintellect"],"value":"Prime Agent looks really interesting. Both the "recursive language model" bit and routing everything through IPython.
https://github.com/PrimeIntellect-ai/prime-agent"},"story_title":{"matchLevel":"none","matchedWords":[],"value":"Quick impressions: A week of using Codex more than Claude"},"story_url":{"matchLevel":"none","matchedWords":[],"value":"https://allaboutcoding.ghinda.com/a-week-of-using-codex-more-than-claude/"}},"_tags":["comment","author_hedgehog","story_49393051"],"author":"hedgehog","children":[49395675],"comment_text":"Prime Agent looks really interesting. Both the "recursive language model" bit and routing everything through IPython.
https://github.com/PrimeIntellect-ai/prime-agent","created_at":"2026-08-21T22:08:58Z","created_at_i":1787350138,"objectID":"49394317","parent_id":49393827,"story_id":49393051,"story_title":"Quick impressions: A week of using Codex more than Claude","story_url":"https://allaboutcoding.ghinda.com/a-week-of-using-codex-more-than-claude/","updated_at":"2026-08-22T02:07:18Z"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"rzk"},"title":{"fullyHighlighted":false,"matchLevel":"partial","matchedWords":["prime","agent"],"value":"Prime Agent: A Self-Improving RLM Agent"},"url":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["prime","agent","primeintellect"],"value":"https://github.com/PrimeIntellect-ai/prime-agent"}},"_tags":["story","author_rzk","story_49339427"],"author":"rzk","created_at":"2026-08-18T00:05:41Z","created_at_i":1787011541,"num_comments":0,"objectID":"49339427","points":4,"story_id":49339427,"title":"Prime Agent: A Self-Improving RLM Agent","updated_at":"2026-08-18T06:48:34Z","url":"https://github.com/PrimeIntellect-ai/prime-agent"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"swills"},"title":{"fullyHighlighted":false,"matchLevel":"partial","matchedWords":["prime","agent"],"value":"Prime Agent: A Self-Improving RLM Agent"},"url":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["prime","agent","primeintellect"],"value":"https://github.com/PrimeIntellect-ai/prime-agent"}},"_tags":["story","author_swills","story_49240910"],"author":"swills","created_at":"2026-08-10T08:27:22Z","created_at_i":1786350442,"num_comments":0,"objectID":"49240910","points":4,"story_id":49240910,"title":"Prime Agent: A Self-Improving RLM Agent","updated_at":"2026-08-10T18:22:39Z","url":"https://github.com/PrimeIntellect-ai/prime-agent"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"thewolfpaul"},"title":{"fullyHighlighted":false,"matchLevel":"partial","matchedWords":["agent"],"value":"A self-improving RLM agent for coding workflows and long-running autonomous task"},"url":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["prime","agent","primeintellect"],"value":"https://github.com/PrimeIntellect-ai/prime-agent"}},"_tags":["story","author_thewolfpaul","story_49214109"],"author":"thewolfpaul","created_at":"2026-08-07T18:02:28Z","created_at_i":1786125748,"num_comments":0,"objectID":"49214109","points":3,"story_id":49214109,"title":"A self-improving RLM agent for coding workflows and long-running autonomous task","updated_at":"2026-08-09T00:05:02Z","url":"https://github.com/PrimeIntellect-ai/prime-agent"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"wertyk"},"title":{"fullyHighlighted":false,"matchLevel":"partial","matchedWords":["prime","agent"],"value":"Prime Agent: A Self-Improving RLM Agent"},"url":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["prime","agent","primeintellect"],"value":"https://github.com/PrimeIntellect-ai/prime-agent"}},"_tags":["story","author_wertyk","story_49193137"],"author":"wertyk","created_at":"2026-08-06T06:20:43Z","created_at_i":1785997243,"num_comments":0,"objectID":"49193137","points":2,"story_id":49193137,"title":"Prime Agent: A Self-Improving RLM Agent","updated_at":"2026-08-09T00:05:02Z","url":"https://github.com/PrimeIntellect-ai/prime-agent"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"filtr12"},"story_text":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["prime","agent","primeintellect"],"value":"This integration allows for scalable evals and training of browser agents with hosted Prime Intellect eval + training pipelines and headless browser infrastructure on Browserbase to RL train browser agents with LoRA."},"title":{"fullyHighlighted":false,"matchLevel":"partial","matchedWords":["agent"],"value":"Show HN: I built an integration for RL training of browser agents for everyone"},"url":{"fullyHighlighted":false,"matchLevel":"partial","matchedWords":["primeintellect"],"value":"https://github.com/PrimeIntellect-ai/verifiers/tree/main/verifiers/envs/integrations/browser_env"}},"_tags":["story","author_filtr12","story_47520234","show_hn"],"author":"filtr12","children":[47520371,47521868,47521998],"created_at":"2026-03-25T17:11:07Z","created_at_i":1774458667,"num_comments":1,"objectID":"47520234","points":7,"story_id":47520234,"story_text":"This integration allows for scalable evals and training of browser agents with hosted Prime Intellect eval + training pipelines and headless browser infrastructure on Browserbase to RL train browser agents with LoRA.","title":"Show HN: I built an integration for RL training of browser agents for everyone","updated_at":"2026-04-07T23:36:27Z","url":"https://github.com/PrimeIntellect-ai/verifiers/tree/main/verifiers/envs/integrations/browser_env"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"Xeophon"},"title":{"fullyHighlighted":false,"matchLevel":"partial","matchedWords":["prime","agent"],"value":"Prime Agent: A self-improving RLM agent"},"url":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["prime","agent","primeintellect"],"value":"https://www.primeintellect.ai/blog/prime-agent"}},"_tags":["story","author_Xeophon","story_49189075"],"author":"Xeophon","children":[49189823,49189863,49190028,49190083,49190153,49190228,49190331,49190852,49191400,49191860,49192115,49192663,49193141,49193303,49194368,49194454,49196223,49197089,49209484],"created_at":"2026-08-05T21:11:57Z","created_at_i":1785964317,"num_comments":69,"objectID":"49189075","points":254,"story_id":49189075,"title":"Prime Agent: A self-improving RLM agent","updated_at":"2026-08-27T16:27:28Z","url":"https://www.primeintellect.ai/blog/prime-agent"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"tosh"},"title":{"fullyHighlighted":true,"matchLevel":"partial","matchedWords":["prime","agent"],"value":"Prime Agent"},"url":{"fullyHighlighted":false,"matchLevel":"partial","matchedWords":["primeintellect"],"value":"https://twitter.com/PrimeIntellect/status/2085086999267144083"}},"_tags":["story","author_tosh","story_49188743"],"author":"tosh","created_at":"2026-08-05T20:44:06Z","created_at_i":1785962646,"num_comments":0,"objectID":"49188743","points":3,"story_id":49188743,"title":"Prime Agent","updated_at":"2026-08-05T22:20:21Z","url":"https://twitter.com/PrimeIntellect/status/2085086999267144083"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"mromanuk"},"title":{"fullyHighlighted":false,"matchLevel":"partial","matchedWords":["prime","agent"],"value":"Prime Agent a general-purpose coding harness. On ARC-AGI-3 scored 95.5%"},"url":{"fullyHighlighted":false,"matchLevel":"partial","matchedWords":["primeintellect"],"value":"https://x.com/PrimeIntellect/thread/2085087002769379520"}},"_tags":["story","author_mromanuk","story_49190818"],"author":"mromanuk","created_at":"2026-08-06T00:18:33Z","created_at_i":1785975513,"num_comments":0,"objectID":"49190818","points":2,"story_id":49190818,"title":"Prime Agent a general-purpose coding harness. On ARC-AGI-3 scored 95.5%","updated_at":"2026-08-06T00:32:06Z","url":"https://x.com/PrimeIntellect/thread/2085087002769379520"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"Danau5tin"},"comment_text":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["prime","agent","primeintellect"],"value":"My RL trained multi-agent-coding model Orca-Agent-v0.1-14B reached a 167% higher relative score than its base model on Stanford's TerminalBench. I've open sourced everything.
*What I did:*
- I trained a 14B orchestrator model to better coordinate explorer & coder subagents (subagents are tool calls for orchestrator)\n- Scaled to 32x H100s that were pushed to their limits across 4 bare-metal nodes\n- Scaled to 256 Docker environments rolling out simultaneously, automatically distributed across the cluster
*Key results:*
- Qwen3-14B jumped from *7% \u2192 18.25%* on TerminalBench after training\n- Model now within striking distance of Qwen3-Coder-480B (19.7%)\n- Training was stable with smooth entropy decrease and healthy gradient norms
*Training approach:*
Reward design and biggest learning: Kept it simple - *just unit tests*. Every "smart" reward signal I tried to craft led to policy collapse
Curriculum learning:\n- Stage-1: Tasks where base model succeeded 1-2/3 times (41 tasks)\n- Stage-2: Tasks where Stage-1 model succeeded 1-4/5 times
Dataset: Used synthetically generated RL environments and unit tests
*More details:*
I have added lots more details in the repo linked to this submission, including training code, model weights, datasets.
Huge thanks to:\n- Tara for providing the compute\n- Prime Intellect team for building prime-rl and dealing with my endless questions \n- Alex Dimakis for the conversation that sparked training the orchestrator model
Thanks for reading!
Dan
(Evaluated on the excellent TerminalBench benchmark by Stanford & Laude Institute)"},"story_title":{"fullyHighlighted":false,"matchLevel":"partial","matchedWords":["agent"],"value":"Scaling Coding-Agent RL to 32x H100s. 160% Improvement on Stanford's TBench"},"story_url":{"fullyHighlighted":false,"matchLevel":"partial","matchedWords":["agent"],"value":"https://github.com/Danau5tin/Orca-Agent-RL"}},"_tags":["comment","author_Danau5tin","story_45798344"],"author":"Danau5tin","comment_text":"My RL trained multi-agent-coding model Orca-Agent-v0.1-14B reached a 167% higher relative score than its base model on Stanford's TerminalBench. I've open sourced everything.
*What I did:*
- I trained a 14B orchestrator model to better coordinate explorer & coder subagents (subagents are tool calls for orchestrator)\n- Scaled to 32x H100s that were pushed to their limits across 4 bare-metal nodes\n- Scaled to 256 Docker environments rolling out simultaneously, automatically distributed across the cluster
*Key results:*
- Qwen3-14B jumped from *7% \u2192 18.25%* on TerminalBench after training\n- Model now within striking distance of Qwen3-Coder-480B (19.7%)\n- Training was stable with smooth entropy decrease and healthy gradient norms
*Training approach:*
Reward design and biggest learning: Kept it simple - *just unit tests*. Every "smart" reward signal I tried to craft led to policy collapse
Curriculum learning:\n- Stage-1: Tasks where base model succeeded 1-2/3 times (41 tasks)\n- Stage-2: Tasks where Stage-1 model succeeded 1-4/5 times
Dataset: Used synthetically generated RL environments and unit tests
*More details:*
I have added lots more details in the repo linked to this submission, including training code, model weights, datasets.
Huge thanks to:\n- Tara for providing the compute\n- Prime Intellect team for building prime-rl and dealing with my endless questions \n- Alex Dimakis for the conversation that sparked training the orchestrator model
Thanks for reading!
Dan
(Evaluated on the excellent TerminalBench benchmark by Stanford & Laude Institute)","created_at":"2025-11-03T12:29:10Z","created_at_i":1762172950,"objectID":"45798345","parent_id":45798344,"story_id":45798344,"story_title":"Scaling Coding-Agent RL to 32x H100s. 160% Improvement on Stanford's TBench","story_url":"https://github.com/Danau5tin/Orca-Agent-RL","updated_at":"2026-03-05T23:01:03Z"}],"hitsPerPage":20,"nbHits":11,"nbPages":1,"page":0,"params":"query=prime+agent+primeintellect&advancedSyntax=true&analyticsTags=backend","processingTimeMS":17,"processingTimingsMS":{"_request":{"queue":106,"roundTrip":15},"fetch":{"query":15,"scanning":1,"total":17},"total":17},"query":"prime agent primeintellect","serverTimeMS":125}