---
format: "aidr-story-markdown/v1"
id: "46375b5b723cecaea5103a1e85bfde764da4553e9105a880d4608dd998d48051"
canonical_url: "https://aidr.today/46375b5b?lang=en"
title: "Octopus: One Brain, Many Branches for Latent-KV Multi-Agent Reinforcement Learning"
lang: "en"
requested_lang: "en"
available_langs: ["en","vi"]
translation_fallback: null
fallback_fields: []
published_at: "2026-11-06T05:00:00.000Z"
category: "Frameworks"
topics: ["agent","framework"]
source_urls: ["https://arxiv.org/abs/2611.06956"]
summary: "Octopus is a single-model multi-agent RL framework. All threads share one backbone and coordinate over a GPU-resident directed acyclic graph (DAG) of KV states. This framework uses five primitives including fork, read_thread, wait_threads, rewind, and commit."
---

# Octopus: One Brain, Many Branches for Latent\-KV Multi\-Agent Reinforcement Learning

> [Open the canonical story](<https://aidr.today/46375b5b?lang=en>)

**Published:** 2026-11-06T05:00:00.000Z
**Category:** Frameworks
**Topics:** agent, framework

## Summary

Octopus is a single\-model multi\-agent RL framework\. All threads share one backbone and coordinate over a GPU\-resident directed acyclic graph \(DAG\) of KV states\. This framework uses five primitives including fork, read\_thread, wait\_threads, rewind, and commit\.

## Sources

- [Story source](<https://arxiv.org/abs/2611.06956>)