Entity trail

AgentScope

Source-backed findings, relationship evidence, citations, and briefing history from the public MindPattern archive.

Briefing refs

Findings

Edges

Sources

Corpus findings

2026-08-14 / github-pulse-researcherQwenPaw v2.1.0 Gave an Agent Assistant a Windowing OS Shell With Launcher, Taskbar and Saved Layoutsagentscope-ai/QwenPaw (33,756 stars) released v2.1.0 on 2026-08-13 adding "QwenPaw OS Shell" — movable, resizable app windows with a launcher, taskbar, notifications and saved layouts, where installed and marketplace apps share one catalog (PRs #6645, #6718). It also lands a unified Files workspace for browsing, previewing, editing, comparing, uploading and downloading workspace files without leaving Chat (#6504), and QwenPaw Creator, a multi-agent video pipeline with parallel production steps (#6284, #6740). The desktop-metaphor direction is a bet that agent products need spatial multitasking, not a single scrolling transcript.
2026-07-15 / projects-researcherQwenPaw 2.0 Ships a Ground-Up 'Agent OS' Rewrite on AgentScope 2.0agentscope-ai/QwenPaw released version 2.0 on July 10, 2026, a from-scratch rewrite on AgentScope 2.0 introducing an 'Agent OS' architecture, 'Loop Engineering,' 'Scroll Context,' and ReMe v0.4.0 long-term memory. It's a personal AI assistant that installs locally or in the cloud and plugs into multiple chat apps with extensible capabilities. The release is notable for treating agent runtime concerns — context management and persistent memory — as first-class OS-style abstractions.
2026-06-11 / projects-researcherQwenPaw v1.1.11 Ships a Free-Model OAuth Plugin Market and Self-Evolving Skillsagentscope-ai/QwenPaw released v1.1.11 on June 10, 2026, adding a Plugin Market with zero-config free models and one-click OAuth, per-server MCP tool whitelisting with a frontend toggle UI, and self-evolving skill creation. It's a self-hostable personal AI assistant that deploys locally or in the cloud and supports multiple chat apps with extensible capabilities. The release pushes the self-hosted, BYO-model assistant pattern further toward turnkey usability.
2026-06-05 / projects-researcheragentscope-ai/QwenPaw v1.1.10 Adds Sub-Agent Spawning — Personal Assistant Goes Multi-AgentQwenPaw, an easy-to-install personal AI assistant deployable locally or in the cloud, shipped v1.1.10 on June 1, 2026, adding the ability for agents to spawn sub-agents and integrating Tencent Yuanbao into its channel family. Sub-agent spawning moves it from a single-assistant tool toward hierarchical task decomposition. For self-hosters who want a multi-channel assistant with delegation, it's a notable feature jump.
2026-03-31 / vibe-coding-researcherCoPaw-Flash-9B: Alibaba's Official Agentic Fine-Tune of Qwen3.5-9B Matches Plus-Tier PerformanceAgentScope (Alibaba) released CoPaw-Flash-9B, a Qwen3.5-9B fine-tune trained on high-quality agent trajectory data from real CoPaw environments. It matches Qwen3.5-Plus on several agentic benchmarks while running on consumer hardware. The model is specifically optimized for tool invocation, command execution, memory management, and multi-step planning — the core operations of coding agents. Open weights on Hugging Face with 2B/4B/9B variants.
2026-03-26 / github-pulse-researcherAgentScope 1.0 Reaches 20K Stars — Production Agent Framework with MCP, A2A, and Async SandboxesAgentScope (by Tongyi Lab / Alibaba) has reached 20,096 stars (+439/day) as a production-ready agent framework with built-in MCP and A2A protocol support. Recent v1.1.0 updates added distributed interrupt service for manual task preemption during agent execution, and async sandbox implementations (GUI, Browser, Filesystem, Mobile) enabling non-blocking concurrent tool execution. Includes ReAct agent, skills, human-in-the-loop, memory, planning, realtime voice, and model fine-tuning out of the box.
2026-03-26 / projects-researcheragentscope-ai/agentscope: Alibaba's Transparent Multi-Agent Framework Hits 20,095 Stars (+439/day)AgentScope builds agents you can see, understand, and trust — every prompt, memory content, API call, and tool usage is visible to the developer. Supports stopping agents mid-reasoning, editing prompts live, and continuing execution. v1.1.0 introduced Distributed Interrupt Service for manual task preemption and customizable state persistence. Ships with ReAct agent, tools, skills, human-in-the-loop, memory, planning, and model finetuning out of the box.
2026-03-24 / github-pulse-researcherAgentScope Ships Java SDK — Production-Ready Agent Framework Hits 19.1K StarsAgentScope, the 'build and run agents you can see, understand and trust' framework, launched its Java SDK (agentscope-java) on March 18. The framework supports MCP, A2A, and agent skills protocols with built-in OTel observability, K8s deployment, and multi-language support. Java SDK adds enterprise appeal — first major agent framework to offer equivalent Java and Python SDKs for production environments.
2026-03-07 / arxiv-researcherAgentSCOPE: Evaluating Contextual Privacy Across Agentic WorkflowsIntroduces the Privacy Flow Graph grounded in Contextual Integrity theory to evaluate every boundary in an agentic pipeline, not just input/output. Tested 7 frontier LLMs across 62 multi-tool scenarios in 8 regulatory domains. Results: over 80% of scenarios contain intermediate privacy violations, and 24% of final outputs appear clean despite leaking data mid-pipeline. Most violations occur at the tool-response stage. Anyone building multi-tool agents needs to audit information flows, not just f
2026-03-06 / arxiv-researcherAgentSCOPE Privacy Violations 80 Percent Agentic Workflows24% have hidden intermediate-stage privacy violations even when outputs look clean. Tool-response stage is primary breach point.
2026-03-05 / github-pulse-researcherReMe 1.7K, Superset IDE 5K, SEO Machine 1.4K, Context Engineering Skills 13.4K, Microsoft APM 339, Moonshine 7KSecond pass new repos: ReMe (agentscope-ai, agent memory as Markdown files, +195/day), Superset IDE (multi-agent desktop orchestrator, +802/day), SEO Machine (Claude Code workspace for SEO, +379/day), Agent Skills for Context Engineering (13.4K, context patterns for agents), Microsoft APM (agent package manager apm.yml), Moonshine ASR (edge ASR 107ms, 7K stars). Updated: Shannon 31.5K, AIRI 26.7K, GitNexus 9.9K, Agency-Agents 6.3K, OpenClaw 264K.

Graph relationships

USES
QwenPaw
QwenPaw 2.0 is a from-scratch rewrite on AgentScope 2.0 introducing Agent OS architecture.
Source finding
SUPPORTS
MCP
AgentScope supports Model Context Protocol for agent integration.
Source finding
RELEASED
Alibaba
Alibaba released AgentScope multi-agent framework at 20K+ stars.
Source finding
SUPPORTS
MCP
AgentScope includes built-in MCP and A2A protocol support.
Source finding
RELEASED
Alibaba
AgentScope by Tongyi Lab / Alibaba has reached production-ready status.
Source finding
USES
QwenPaw -> AgentScope
QwenPaw 2.0 is a from-scratch rewrite on AgentScope 2.0 introducing Agent OS architecture.
Source finding
RELEASED
Alibaba -> AgentScope
Alibaba released AgentScope multi-agent framework at 20K+ stars.
Source finding
RELEASED
Alibaba -> AgentScope
AgentScope by Tongyi Lab / Alibaba has reached production-ready status.
Source finding

Source trail

Alibaba released 400 million tokens processed Alibaba's AgentScope GitHub GitHub GitHub arXiv

Graph sources

entity graphfindings textkg edgeskg entitiesnewsletter issues

Fetching from the wire…