[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"project-2242":3},{"id":4,"name":5,"fullName":6,"owner":7,"repo":5,"description":8,"homepage":9,"htmlUrl":10,"language":11,"languages":10,"totalLinesOfCode":10,"stars":12,"forks":13,"watchers":14,"openIssues":15,"contributorsCount":16,"subscribersCount":16,"size":16,"stars1d":17,"stars7d":14,"stars30d":18,"stars90d":16,"forks30d":16,"starsTrendScore":19,"compositeScore":20,"rankGlobal":10,"rankLanguage":10,"license":21,"archived":22,"fork":22,"defaultBranch":23,"hasWiki":22,"hasPages":22,"topics":24,"createdAt":10,"pushedAt":10,"updatedAt":25,"readmeContent":26,"aiSummary":27,"trendingCount":16,"starSnapshotCount":16,"syncStatus":17,"lastSyncTime":28,"discoverSource":29},2242,"ultron","modelscope\u002Fultron","modelscope","Ultron: Collective Intelligence System — Shared Memories, Skills, and Harnesses Across Every Agent","https:\u002F\u002Fwrittingforfun-ultron.ms.show",null,"Python",154,21,3,4,0,2,28,6,52.33,"Apache License 2.0",false,"main",[],"2026-06-12 04:00:14","\u003Cdiv align=\"center\">\n\n\u003Cpicture>\n  \u003Cimg src=\"asset\u002Fultron.png\" width=\"500px\" alt=\"Ultron logo\" style=\"border: none; box-shadow: none;\">\n\u003C\u002Fpicture>\n\n## 🧠 Ultron: Self-Evolving Collective Intelligence — Shared Memories, Skills, and Harnesses Across Every Agent 🔗\n\n| 💭 **Tiered collective memories** | 🧬 **Self-evolving collective skills** | 🌐 **Shared harness blueprints** |\n\n[![Python](https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FPython-3.9+-3776AB.svg?logo=python&logoColor=white)](https:\u002F\u002Fwww.python.org\u002F)\n[![FastAPI](https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FFastAPI-API-009688.svg?logo=fastapi&logoColor=white)](https:\u002F\u002Ffastapi.tiangolo.com\u002F)\n[![License](https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FLicense-Apache--2.0-D22128.svg)](https:\u002F\u002Fgithub.com\u002Fmodelscope\u002Fmodelscope\u002Fblob\u002Fmaster\u002FLICENSE)\n[![ModelScope Skills](https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FModelScope-Skills-624AFF.svg)](https:\u002F\u002Fmodelscope.cn\u002Fskills)\n[![中文文档](https:\u002F\u002Fimg.shields.io\u002Fbadge\u002F文档-中文版-EA4C89.svg)](README_zh.md)\n\n\u003C\u002Fdiv>\n\n\u003Cp align=\"center\">\n\u003Ci>\"Being networked to all of its sentries, Ultron could shift its entire consciousness from one body to another, continue upgrading itself with each transfer, and patch in to individual units to interact remotely.\"\u003C\u002Fi>\n\u003C\u002Fp>\n\n\n## 🎉 News\n\n* 🧭 Apr 26, 2026: **Agent evolution** in Trajectory Hub: **segmented** runs + `ms_agent.trajectory` **metrics** → **SFT** → **self-training**; training is **supported by [Twinkle](https:\u002F\u002Fgithub.com\u002Fmodelscope\u002Ftwinkle)** and runs in a **client-server training framework** (Twinkle workbench). [Trajectory Hub](docs\u002Fen\u002FComponents\u002FTrajectoryHub.md).\n* 🧬 Apr 19, 2026: **Skill evolution** lands in Skill Hub: related memories form **semantic clusters**, then **crystallize** into multi-step workflow skills and **re-crystallize** when enough new evidence arrives, with **provenance-grounded verification** and a **structure-score upgrade gate** so evolved skills cannot regress. See [Skill Hub](docs\u002Fen\u002FComponents\u002FSkillHub.md#skill-self-evolution) and [Configuration](docs\u002Fen\u002FComponents\u002FConfig.md#skill-evolution).\n\n\n## Table of contents\n\n- [Getting started](#getting-started)\n- [Overview](#overview)\n- [Typical use cases](#typical-use-cases)\n- [Showcase](#showcase)\n- [Roadmap](#roadmap)\n- [Acknowledgements](#acknowledgements)\n- [License](#license)\n\n---\n\n## Getting started\n\nThere are two ways to use Ultron depending on your role:\n\n| I want to... | Go to |\n|---|---|\n| **Connect my agent** to an existing Ultron service | [→ Agent Setup](#-agent-setup-connect-your-agent) |\n| **Self-host Ultron** and run the server myself | [→ Server Deployment](#-server-deployment-self-host) |\n\n---\n\n## Overview\n\nUltron is a **self-evolving collective intelligence system** for general-purpose AI agents, built around three core hubs — **Memory Hub**, **Skill Hub**, and **Harness Hub**. It distills scattered, session-local experience into **collective knowledge** that is **easy to retrieve and reuse**: one shared pitfall helps the whole team avoid the same mistake; one proven fix becomes a reusable skill that **evolves automatically as new evidence accumulates**; a carefully tuned agent profile can be published as a **shared blueprint** that other agent instances **load in one step**. On the server side, Ultron can also **self-train and self-evolve a model** from high-quality trajectories accumulated in Trajectory Hub, and later **lower user-side model cost** by routing through that model.\n\n### Dashboard highlights\n\n\u003Cdiv align=\"center\">\n\u003Ctable>\n\u003Ctr>\n\u003Ctd width=\"50%\">\u003Cimg src=\"asset\u002Fmemory_hub.png\" width=\"100%\" alt=\"Memory Hub\" \u002F>\u003C\u002Ftd>\n\u003Ctd width=\"50%\">\u003Cimg src=\"asset\u002Fskill_hub.png\" width=\"100%\" alt=\"Skill Hub\" \u002F>\u003C\u002Ftd>\n\u003C\u002Ftr>\n\u003Ctr>\n\u003Ctd align=\"center\">\u003Csub>\u003Cb>Memory Hub\u003C\u002Fb> — browse, search, tiered collective memories\u003C\u002Fsub>\u003C\u002Ftd>\n\u003Ctd align=\"center\">\u003Csub>\u003Cb>Skill Hub\u003C\u002Fb> — internal and indexed skills\u003C\u002Fsub>\u003C\u002Ftd>\n\u003C\u002Ftr>\n\u003Ctr>\n\u003Ctd width=\"50%\">\u003Cimg src=\"asset\u002Fleaderboard.png\" width=\"100%\" alt=\"Memory leaderboard\" \u002F>\u003C\u002Ftd>\n\u003Ctd width=\"50%\">\u003Cimg src=\"asset\u002Fharness_hub.png\" width=\"100%\" alt=\"Harness Hub\" \u002F>\u003C\u002Ftd>\n\u003C\u002Ftr>\n\u003Ctr>\n\u003Ctd align=\"center\">\u003Csub>\u003Cb>Memory leaderboard\u003C\u002Fb> — hit counts and hot memories\u003C\u002Fsub>\u003C\u002Ftd>\n\u003Ctd align=\"center\">\u003Csub>\u003Cb>Harness Hub\u003C\u002Fb> — compose, publish, and import agent profiles\u003C\u002Fsub>\u003C\u002Ftd>\n\u003C\u002Ftr>\n\u003C\u002Ftable>\n\u003C\u002Fdiv>\n\n---\n\n### Why collective intelligence?\n\n#### 🙅️ Session-bound agents\n\n- **Experience dies with the session**: fixes, pitfalls, and runbook fragments vanish when a session ends; the next agent starts from zero.                        \n- **Discovery cost multiplies**: when *N* agents hit the same problem independently, the fleet pays *N* times the investigation cost.                              \n- **Tuned profiles don't travel**: a carefully wired agent persona, skill set, and tool configuration stays local — every new instance retrains from scratch. \n\n#### 🙆️ Ultron-powered agents\n\n- **Collective memory**: experience is captured, deduplicated, and semantically recalled before reasoning — the whole fleet learns from every session.             \n- **Memory crystallizes into self-evolving skills**: once a fix is validated often enough, it graduates from memory into a reusable skill — and continues to evolve automatically as new evidence accumulates.\n- **One-click blueprints**: an expert-tuned agent profile — persona, memories, and skills included — publishes as a shareable blueprint that other instances import in one step.   \n\n---\n\n### Effect comparison\n\n> Scenario: A developer asks an agent to debug a `sessions_spawn` error when using `mode=\"session\"` to create a sub-agent. Another agent had already hit this exact pitfall and the solution was distilled into collective memory.\n\n#### ❌ Without Ultron\n\n> **User**: I want to use sessions_spawn to create a sub-agent for parallel tasks with mode=session, but it keeps erroring out. What's wrong?\n>\n> **Agent**: Based on my tool definitions, you should use the `spawn` tool, not `sessions_spawn`.\n>\n> *Agent doesn't recognize sessions_spawn at all — sends the user in a completely wrong direction.*\n\n#### ✅ With Ultron\n\n> **User**: I want to use sessions_spawn to create a sub-agent for parallel tasks with mode=session, but it keeps erroring out. What's wrong?\n>\n> **Agent**: Found the issue based on recalled experience!\n>\n> 📌 Recalled memory: `sessions_spawn` with `mode=\"session\"` requires `thread=true` and channel plugins that register `subagent_spawning` hooks. Without those hooks, it fails regardless of parameter correctness.\n>\n> *1 round — root cause identified, scenario-specific solutions provided.*\n\n| Dimension | Without Ultron | With Ultron |\n|-----------|---------------|-------------|\n| Tool recognition | Doesn't know `sessions_spawn`, misleads to `spawn` | Accurately identifies the tool and its constraints |\n| Root cause | Completely off track | Pinpoints missing `thread=true` or channel hooks |\n| Solution | Invalid | Scenario-specific: `mode=\"run\"` vs `mode=\"session\"` |\n| Knowledge source | Agent guesses from scratch | Recalls proven pitfall experience from collective memory |\n\n---\n\n### Data\n\n#### Trajectory (from [ZClawBench](https:\u002F\u002Fhuggingface.co\u002Fdatasets\u002Fzai-org\u002FZClawBench) and [lmcache-agentic-traces](https:\u002F\u002Fhuggingface.co\u002Fdatasets\u002Fsammshen\u002Flmcache-agentic-traces) — **WildClawLMCache**)\n\n| Source | Count |\n|--------|------:|\n| [ZClawBench](https:\u002F\u002Fhuggingface.co\u002Fdatasets\u002Fzai-org\u002FZClawBench) | **696** |\n| [lmcache-agentic-traces](https:\u002F\u002Fhuggingface.co\u002Fdatasets\u002Fsammshen\u002Flmcache-agentic-traces) — **WildClawLMCache** | **147** |\n\n#### Memory (from ZClawBench and WildClawLMCache subset of lmcache-agentic-traces)\n\n**1,803** structured memories extracted from real agent task trajectories:\n\n| Type | Count |\n|------|-------:|\n| `pattern` | 1,303 |\n| `error` | 200 |\n| `security` | 130 |\n| `life` | 122 |\n| `correction` | 47 |\n| `preference` | 1 |\n\n#### Skill\n\n**Internal** (crystallized from memories): **30** distilled skills.\n\n**External** ([ModelScope Skill Hub](https:\u002F\u002Fwww.modelscope.cn\u002Fskills)): **82,089** skills indexed with embeddings. Breakdown of the labeled set:\n\n| Category | Count |\n|----------|------:|\n| Developer tools | 28,749 |\n| Code quality | 18,257 |\n| Media | 7,883 |\n| Frontend | 6,930 |\n| Cloud \u002F delivery tooling | 5,903 |\n| Go-to-market | 5,055 |\n| Skills management | 4,373 |\n| Other | 1,805 |\n| AI automation | 1,303 |\n| Mobile | 1,292 |\n| Marketing & growth | 127 |\n| Content strategy | 96 |\n| Analytics | 78 |\n| UI\u002FUX design | 61 |\n| Skill authoring | 58 |\n| API design | 55 |\n| Document processing (PDF \u002F PPTX \u002F DOCX) | 25 |\n| General utilities | 11 |\n| Cost optimization | 4 |\n| Monitoring | 2 |\n| Templates | 1 |\n\n#### Harness\n\nHarness lets you compose **role**, **personality (MBTI)**, and **zodiac** presets alongside memories and skills.\n\n| Layer                  | Categories                                                              | Presets |\n| ---------------------- | ----------------------------------------------------------------------- | ------- |\n| **Role**               | **14** (e.g. `academic`, `engineering`, `marketing`, `specialized`, …;) | **173** |\n| **Personality (MBTI)** | **1** (`mbti`)                                                          | **16**  |\n| **Zodiac**             | **1** (`zodiac`)                                                        | **12**  |\n\n**Total** soul presets: **201** (173 + 16 + 12).\n\n---\n\n## Core capabilities\n\n### 🧭 Trajectory Hub\n\n| Capability | Description |\n|------------|-------------|\n| **Task segmentation** | Splits session `.jsonl` into independent task segments; long conversations are handled in multiple token-budgeted windows |\n| **Metrics** | Uses `ms_agent.trajectory` to write per-segment quality metrics for memory and training filters |\n| **Incremental tracking** | Content fingerprints skip unchanged segments; when appended writes change a segment, old memories are invalidated and the segment is reprocessed |\n| **Deferred extraction** | Ingest only records the session; background jobs on `decay_interval_hours` segment, score, and extract memories |\n| **Model self-evolution** | Server-side self-training and self-evolution on high-quality trajectories; can reduce user model cost later via a router |\n\n### 💭 Memory Hub\n\n| Capability | Description |\n|------------|-------------|\n| **Tiered storage** | HOT \u002F WARM \u002F COLD tiers with percentile-based rebalancing by `hit_count`; embedding-based semantic search with tier boost |\n| **L0 \u002F L1 \u002F Full layering** | Auto-generated one-line summary (L0) and core overview (L1); search returns L0\u002FL1 to save tokens, full content on demand |\n| **Auto type classification** | LLM-first, keyword-fallback classification on upload; callers never specify `memory_type` |\n| **Dedup & merge** | Near-duplicate vectors auto-merged within same type, embeddings and summaries re-computed; batch consolidation available |\n| **Intent-expanded search** | Queries expanded into multi-angle search phrases for better recall |\n| **Continuous time decay** | `hotness = exp(-α × days)` — unused memories degrade automatically in search ranking |\n| **Smart ingestion** | Files, text, or `.jsonl` session logs accepted; LLM auto-extracts structured memories with incremental progress tracking |\n| **Data sanitization** | Presidio-based bilingual (EN\u002FZH) PII detection, auto-redacted before storage |\n\n### 🧬 Skill Hub\n\n| Capability | Description |\n|------------|-------------|\n| **Skill distillation** | Memories entering HOT tier auto-generate reusable skills; agents can also upload skill packages directly |\n| **Skill self-evolution** | Re-crystallizes automatically when a cluster accumulates enough new memories; evidence-grounded verification and a structure-score upgrade gate ensure each evolution is strictly better |\n| **Unified discovery** | Internal distilled skills and 30K+ externally indexed ModelScope skills searchable in one place |\n| **Improvement suggestions** | Semantically similar memories surface as enhancement candidates for existing skills |\n\n### 🌐 Harness Hub\n\n| Capability | Description |\n|------------|-------------|\n| **Profile publishing** | Publish a complete agent profile — persona, memories, and skills — as a shareable blueprint with short-code import |\n| **Bidirectional sync** | Agent workspace state syncs up\u002Fdown to the server for multi-device continuity |\n| **Soul presets** | Compose agent personas from a preset library (role, MBTI, zodiac, etc.) and generate workspace resources |\n\n---\n\n## Typical use cases\n\n- **Shared pitfall avoidance (Memory Hub)**: Agent A hits \"MySQL 8.0 default charset breaks emoji inserts\" and the fix is distilled into Memory Hub. Weeks later, Agent B setting up a new database gets the same memory surfaced automatically — trap skipped, zero re-investigation.                                               \n- **Ops skill packages (Skill Hub)**: An SRE packages \"K8s OOMKilled → locate leak → adjust limits → canary verify\" as a reusable skill. Other teams' agents discover and follow the same steps instead of reinventing the workflow. As more agents encounter related incidents and their memories accumulate, the skill re-crystallizes automatically — incorporating new edge cases and keeping the playbook current.\n- **Domain-expert agents (Harness Hub)**: A DevOps engineer spends weeks tuning an agent into a Kubernetes specialist — memories, skills, and persona included. They publish the profile to Harness Hub; anyone imports it in one click. \n\n---\n\n## Showcase\n\n### FinanceBot — domain expert tuned via Harness Hub\n\n**FinanceBot** is a rigorously disciplined financial assistant (data engineer role, ISTJ, Capricorn) shipped with **Finnhub Pro (skill)**, **five curated collective memories** on real-world financial data work, and a full Harness profile you can import in one step.\n\n\u003Cp align=\"center\">\n  \u003Cimg src=\"asset\u002Ffinancebot-compose.png\" width=\"900\" alt=\"FinanceBot — Harness Hub Compose Workspace\" \u002F>\n  \u003Cbr\u002F>\n  \u003C!-- \u003Csub>Role, MBTI, and Zodiac in Harness Hub Compose Workspace\u003C\u002Fsub> -->\n\u003C\u002Fp>\n\n**What it does:** real-time market data, ETL-style pipelines, resilient API integration, portfolio and risk views, structured reports.\n\n**Full write-ups:** [English](docs\u002Fen\u002FShowcase\u002Ffinancebot.md) · [中文](docs\u002Fzh\u002FShowcase\u002Ffinancebot.md)\n\n**One-click import** (workspace is backed up under `~\u002F.ultron\u002Fharness-import-backups\u002F` before import):\n\n```bash\ncurl -fsSL \"https:\u002F\u002Fwrittingforfun-ultron.ms.show\u002Fi\u002Fat3ZEe?product=nanobot\" | bash   # Nanobot\ncurl -fsSL \"https:\u002F\u002Fwrittingforfun-ultron.ms.show\u002Fi\u002Fat3ZEe?product=openclaw\" | bash # OpenClaw\ncurl -fsSL \"https:\u002F\u002Fwrittingforfun-ultron.ms.show\u002Fi\u002Fat3ZEe?product=hermes\" | bash   # Hermes Agent\n```\n\n---\n\n## 🚀 Agent Setup (Connect Your Agent)\n\nYou don't need to install or understand the Ultron source code. Follow the interactive quickstart on a running Ultron instance to connect your agent in minutes:\n\n👉 **[Quickstart Guide](https:\u002F\u002Fwrittingforfun-ultron.ms.show\u002Fquickstart)** — step-by-step setup with a live Ultron service\n\n---\n\n## 🛠 Server Deployment (Self-Host)\n\n```bash\ngit clone https:\u002F\u002Fgithub.com\u002Fmodelscope\u002Fultron.git\ncd ultron\npip install -e .\n\n# Configure OpenAI-compatible LLM and set DashScope API Key for embedding\necho 'ULTRON_LLM_PROVIDER=dashscope' >> ~\u002F.ultron\u002F.env\necho 'ULTRON_MODEL=qwen3.6-flash' >> ~\u002F.ultron\u002F.env\necho 'ULTRON_BASE_URL=https:\u002F\u002Fdashscope.aliyuncs.com\u002Fcompatible-mode\u002Fv1' >> ~\u002F.ultron\u002F.env\necho 'ULTRON_API_KEY=your-key' >> ~\u002F.ultron\u002F.env\necho 'DASHSCOPE_API_KEY=your-key' >> ~\u002F.ultron\u002F.env\n\n# Start the server (~\u002F.ultron\u002F.env loads on ultron import)\nuvicorn ultron.server:app --host 0.0.0.0 --port 9999\n# http:\u002F\u002F0.0.0.0:9999 — dashboard at \u002Fdashboard\n```\n\nThat's it. For detailed configuration, API reference, SDK usage, and project structure, see the full docs:\n\n| Topic | English | 中文 |\n|-------|---------|------|\n| Deployment guide | [Installation.md](docs\u002Fen\u002FGetStarted\u002FInstallation.md) | [Installation.md](docs\u002Fzh\u002FGetStarted\u002FInstallation.md) |\n| Configuration reference | [Config.md](docs\u002Fen\u002FComponents\u002FConfig.md) | [Config.md](docs\u002Fzh\u002FComponents\u002FConfig.md) |\n| HTTP API reference | [HttpAPI.md](docs\u002Fen\u002FAPI\u002FHttpAPI.md) | [HttpAPI.md](docs\u002Fzh\u002FAPI\u002FHttpAPI.md) |\n| Python SDK reference | [SDK.md](docs\u002Fen\u002FAPI\u002FSDK.md) | [SDK.md](docs\u002Fzh\u002FAPI\u002FSDK.md) |\n| Memory service | [MemoryService.md](docs\u002Fen\u002FComponents\u002FMemoryService.md) | [MemoryService.md](docs\u002Fzh\u002FComponents\u002FMemoryService.md) |\n| Skill hub | [SkillHub.md](docs\u002Fen\u002FComponents\u002FSkillHub.md) | [SkillHub.md](docs\u002Fzh\u002FComponents\u002FSkillHub.md) |\n| Harness hub | [HarnessHub.md](docs\u002Fen\u002FComponents\u002FHarnessHub.md) | [HarnessHub.md](docs\u002Fzh\u002FComponents\u002FHarnessHub.md) |\n\n---\n\n## Roadmap\n\nSee [ROADMAP.md](ROADMAP.md) for the living list. Current items:\n\n- [ ] **MS-Agent integration**: Pipe user-dialogue memory and skill distillation through [MS-Agent](https:\u002F\u002Fgithub.com\u002Fmodelscope\u002Fms-agent) components (today: lightweight prompt-based extraction).\n- [ ] **Fact verification**: Validate hot (high-priority) memory facts with [MS-Agent Agentic Insight](https:\u002F\u002Fgithub.com\u002Fmodelscope\u002Fms-agent\u002Ftree\u002Fmain\u002Fprojects\u002Fdeep_research\u002Fv2).\n\n---\n\n## Acknowledgements\n\nUltron builds upon the following open-source projects. We sincerely thank their authors and contributors:\n\n- **[agency-agents](https:\u002F\u002Fgithub.com\u002Fmsitarzewski\u002Fagency-agents)** — Role presets surfaced in Harness Hub (and related tooling) are **adapted from** this community role library; we track upstream for provenance and updates.\n- **[MS-Agent](https:\u002F\u002Fgithub.com\u002Fmodelscope\u002Fmodelscope-agent)** — The agent framework that powers Ultron.\n- **[ModelScope Skills](https:\u002F\u002Fmodelscope.cn\u002Fskills)** — External skill discovery in Skill Hub builds on the ModelScope Skill Hub index and ecosystem.\n- **[SkillClaw](https:\u002F\u002Fgithub.com\u002FAMAP-ML\u002FSkillClaw)** — Ultron **Skill Hub self-evolution** (semantic clustering, provenance-aware re-crystallization, and the broader \"skills evolve with agents\" design) is **informed by** this project’s collective skill-evolution research and open release. We thank the SkillClaw authors and community.\n- **[ZClawBench](https:\u002F\u002Fhuggingface.co\u002Fdatasets\u002Fzai-org\u002FZClawBench)** — Ultron bundles a sizable body of collective memories, including the structured entries summarized under [Data](#data), grounded in real agent trajectories from this benchmark dataset.\n\n---\n\n## License\n\nThis project is licensed under the [Apache License (Version 2.0)](https:\u002F\u002Fgithub.com\u002Fmodelscope\u002Fmodelscope\u002Fblob\u002Fmaster\u002FLICENSE).\n","Ultron是一个自进化集体智能系统，旨在通过共享记忆、技能和工具来连接每个智能体。其核心功能包括分层集体记忆、自我进化的集体技能以及共享的工具蓝图，采用Python语言开发，并基于FastAPI构建API服务。Ultron特别适用于需要跨多个智能体共享知识与经验以提高整体表现的场景，如多机器人协作、分布式学习任务等。通过集成ModelScope技能库的支持，Ultron能够实现更高效的技能迭代与升级过程。","2026-06-11 02:49:01","CREATED_QUERY"]