{"slug": "penguinharness-open-efficient-self-improving-harness", "title": "PenguinHarness, Open, Efficient, Self-Improving Harness", "summary": "PenguinHarness, an open-source agent framework from Prism Shadow, claims to build agents at 100× the speed of LangChain with a zero-code CLI and Web UI connected to 1000+ models, achieving best accuracy on data analysis at 1/70 of Claude Code's cost. The system uses a minimal toolset tuned for open models like DeepSeek, and its Skills enable agents to self-optimize by running benchmarks and shipping improved versions. A complete RAG app can be generated for $0.02 in tokens on DeepSeek V4 Pro.", "body_md": "**With LangChain, you build agents by hand — at 1× speed.With PenguinHarness, agents build agents — at 100×.**\n\nA zero-code Harness CLI and Web UI, connected to 1000+ models.\n\nEnglish | [简体中文](/Prism-Shadow/penguin-harness/blob/main/README.zh.md)\n\nThree reasons, in deliberate order — from task quality, to how agents get built, to how they keep improving.\n\nA deliberately minimal toolset over clean low-level interfaces: fewer tool calls, fewer tokens — deeply tuned for open models like DeepSeek. Each harness on the model it is normally paired with, same tasks, head-to-head:\n\n**Best accuracy on data analysis — at 1/70 of Claude Code's cost.**\n\nType one sentence, and an Agent builds the complete Agent application for you — scaffold, code, and run instructions, end to end:\n\n```\nCollect the docs from https://github.com/ericbuess/claude-code-docs and build a RAG app that answers Claude Code questions as a configuration expert, citing its sources.\n```\n\nAnd this is the finished product — a docs expert with retrieval, cited sources that link to the original files, and example questions built in:\n\n## rag_en.mp4\n\n**And generating this entire RAG app burned just $0.02 (¥0.2) of tokens — on DeepSeek V4 Pro.**\n\nWith PenguinHarness Skills, an Agent evaluates and optimizes itself: run the benchmark, find the lost points, ship version N+1 — with a snapshot before every round, and every request observable in the Trace view.\n\n## evo_en.mp4\n\nFour Skill groups ship in the box ([docs](https://penguin.ooo/docs/skills)); Agents can also write and optimize their own:\n\n| Group | Skills |\n|---|---|\n| Office Productivity | `data-analysis` , `firecrawl` |\n| Software Development | `web-design` , `software-engineering` |\n| AI App Development | `penguin-sdk` , `penguin-cli` , `agenthub-models` , `vllm` , `ollama` , `llamafactory` |\n| Agent Tuning | `agent-creation` , `benchmark-design` , `agent-evaluation` , `agent-optimization` |\n\n| Model | Providers |\n|---|---|\n| DeepSeek V4 | DeepSeek, OpenRouter, Fireworks AI, SiliconFlow, Qwen Token Plan |\n| Kimi K3 | Moonshot AI, OpenRouter, Qwen Pay-As-You-Go |\n| GLM 5.2 | Z.AI, OpenRouter, Fireworks AI, SiliconFlow, Qwen Token Plan, Qwen Pay-As-You-Go |\n| Hunyuan 3 | OpenRouter |\n| Qwen 3.8 Max | Qwen Token Plan (preview) |\n| GPT 5.6 | OpenRouter |\n| Gemini 3.6 Flash | Google Gemini, OpenRouter |\n| Claude 5 | Anthropic, OpenRouter |\n\nEach family's latest generation only — the app's **Models** page lists every built-in preset, and any OpenAI-protocol endpoint works too: pick a preset, or point a custom endpoint at any of the 1000+ online and local models.\n\n| Requirement | Supported |\n|---|---|\n| OS | Linux, macOS |\n| Architecture | x64, arm64 |\n| Runtime | bundled by the one-line installer (npm installs need Node >= 24) |\n| Model | an API key for at least one model |\n\n🚀 Install and launch the full experience (multi-session chat, Agent/skill/model management, usage stats, Trace observability, evaluation center):\n\n```\ncurl -fsSL https://penguin.ooo/install.sh | sh\npenguin web        # start the service and open http://127.0.0.1:7364 (first login: admin / penguin-2026)\n```\n\n📦 Or via npm: `npm install -g @prismshadow/penguin-cli`\n\n. Configure models on the in-app Models page, then chat.\n\nThe same engine, scriptable — made to be driven by agents (and agents building agents):\n\n```\npenguin config model add --provider deepseek --model-id deepseek-v4-pro --api-key sk-... --set-default\npenguin run -m \"Create hello.txt containing Hello, Penguin\"   # one-shot task\npenguin chat       # interactive REPL (/compact, /exit, Ctrl-C to interrupt)\npenguin server     # headless service (same API the Web App uses)\njs\nimport { createAgent, isCompleteModelMessage, userText } from \"@prismshadow/penguin-core\";\n\nconst agent = await createAgent({ agentId: \"default_agent\" });\nconst session = await agent.createSession({ workspaceDir: process.cwd() });\n\nfor await (const output of session.run([userText(\"Create hello.txt containing hi\")], {\n  approve: async () => \"allow\", // per-tool-call approval\n})) {\n  if (isCompleteModelMessage(output) && output.payload.type === \"text\") {\n    console.log(output.payload.text);\n  }\n}\n```\n\n- Public release of the benchmark suite\n- Desktop app\n- Windows support\n- Agent company and templates\n- Company-level self evolving\n- OpenShell integration (permission-governed shell)\n- More to come…\n\n```\npnpm install && pnpm build   # build first: core's exports point at dist/\npnpm dev                     # backend + web app together (prefixed logs, deps built once)\n```\n\nSee [CONTRIBUTING.md](/Prism-Shadow/penguin-harness/blob/main/CONTRIBUTING.md) for the full workspace guide: dev commands, quality gates, repo layout, and the changelog rule.\n\nIf you use PenguinHarness in your research, please cite:\n\n```\n@software{penguinharness2026,\n  author  = {{PrismShadow Team}},\n  title   = {PenguinHarness: Efficient Self-Improving Harness for Everyone},\n  year    = {2026},\n  url     = {https://github.com/Prism-Shadow/penguin-harness},\n  license = {Apache-2.0}\n}\n```\n\n[Apache-2.0](/Prism-Shadow/penguin-harness/blob/main/LICENSE) © 2026 Prism Shadow\n\nBuilt with ❤️ by [Yaowei Zheng](https://github.com/hiyouga) (author of [LlamaFactory](https://github.com/hiyouga/LlamaFactory)), the [PrismShadow AI Team](https://github.com/Prism-Shadow), and [Fable 5](https://www.anthropic.com/news/claude-fable-5-mythos-5).", "url": "https://wpnews.pro/news/penguinharness-open-efficient-self-improving-harness", "canonical_source": "https://github.com/Prism-Shadow/penguin-harness", "published_at": "2026-07-23 00:14:20+00:00", "updated_at": "2026-07-23 00:52:53.175020+00:00", "lang": "en", "topics": ["ai-agents", "ai-tools", "ai-infrastructure", "developer-tools", "large-language-models"], "entities": ["PenguinHarness", "Prism Shadow", "LangChain", "DeepSeek", "Claude Code", "DeepSeek V4 Pro", "OpenRouter", "Fireworks AI"], "alternates": {"html": "https://wpnews.pro/news/penguinharness-open-efficient-self-improving-harness", "markdown": "https://wpnews.pro/news/penguinharness-open-efficient-self-improving-harness.md", "text": "https://wpnews.pro/news/penguinharness-open-efficient-self-improving-harness.txt", "jsonld": "https://wpnews.pro/news/penguinharness-open-efficient-self-improving-harness.jsonld"}}