--- title: DeepSeek releases open-source agent harness with 150,000... description: DeepSeek released an open-source, MIT-licensed agent harness called dsh that gained 150,000 GitHub stars within days. Built on the Cordis plugin framework with... canonical: https://daily.dev/posts/deepseek-releases-open-source-agent-harness-with-150-000-github-stars-in-days-wi2yf4gtc twitter:card: summary_large_image twitter:site: @dailydotdev og:type: website og:site_name: daily.dev og:title: DeepSeek releases open-source agent harness with 150,000 GitHub stars in days | daily.dev og:description: DeepSeek released an open-source, MIT-licensed agent harness called dsh that gained 150,000 GitHub stars within days. Built on the Cordis plugin framework with... og:url: https://daily.dev/posts/deepseek-releases-open-source-agent-harness-with-150-000-github-stars-in-days-wi2yf4gtc og:image: https://api.daily.dev/og/posts/wI2YF4gTC.png og:image:alt: DeepSeek releases open-source agent harness with 150,000 GitHub stars in days og:image:width: 1200 og:image:height: 630 og:locale: en --- > ## Documentation Index > Fetch the complete documentation index at: https://daily.dev/llms.txt > Use this file to discover all available pages before exploring further. # DeepSeek releases open-source agent harness with 150,000 GitHub stars in days **[Collections](https://daily.dev/sources/collections)** · 3 min read · 3 upvotes · 0 comments ## Summary DeepSeek released an open-source, MIT-licensed agent harness called dsh that gained 150,000 GitHub stars within days. Built on the Cordis plugin framework with a modular micro-kernel architecture, it lets users swap in any model backend (Claude Sonnet 4.5, OpenAI, Bedrock, etc.) rather than being locked to DeepSeek's own models. Version 0.1 ships with four modes - Standard, Code, Minimal, and Creator - with Creator allowing plugin authoring purely through prompts. It also logs every prompt, reasoning step, and tool call to an append-only trace for debugging. The community quickly built a desktop app, TUI, web UI skins, and an awesome-list. DeepSeek also released V4 Pro, priced below Opus and reportedly matching GLM 5.2 on intelligence benchmarks. ## Content DeepSeek released an open-source agent harness called DeepSeek Harness (dsh), built on top of an existing plugin framework called Cordis, and it hit 202,000 GitHub stars within days. That number is worth pausing on - this isn't a polished consumer product, it's a developer preview with warned breaking changes, and the community showed up anyway. ## What it actually is The core idea is that every component is a swappable plugin: models, tools, skills, memory, execution loops, permissions. Nothing is hardcoded. You can point it at DeepSeek's own API, GLM, Gemini, a local Ollama instance, Claude Sonnet, OpenAI, or Bedrock - the harness doesn't care. This is the meaningful difference from something like Claude Code, which bundles a fixed set of decisions into a closed product you can't inspect or modify. It ships with four modes out of the box: Standard, Code, Minimal, and Creator. Creator mode is the interesting one - it lets you author new plugins through prompting rather than writing code directly. The demo showed someone adding a playable dinosaur game to the desktop app just by asking. More practically, people have used it to generate code review modes, research fact-checking workflows, GPU monitoring dashboards, and storyboarding tools. ## The self-modification angle An 88-page paper accompanies the release explaining how the system handles self-modification without falling apart. Every change the agent makes to its own interface or behavior comes with automatic cleanup and undo instructions baked in, so modifications are reversible. That's not a small engineering problem - most agent systems that can rewrite themselves tend to drift into incoherence. Whether this approach holds up at scale is still an open question, but the design is at least thoughtful about it. ## Traceability and resumability Two features stand out for anyone who's been burned by opaque agent runs. First, append-only traceability logs every system prompt, reasoning step, and tool call - you can see exactly what happened and why. Second, killed tasks can be resumed from their exact context rather than restarted from scratch. Both of these feel obvious in retrospect, but most agent tooling still doesn't do them well. The harness also ships sub-agent packages that can spawn actual Claude Code or Codex processes as callable tools, which is a useful escape hatch when you want the harness's flexibility but need a specific tool's capabilities. ## Community response Within days of release, the community built a desktop app (11,000 stars on its own), a TUI, several web UI skins, and an awesome-list of plugins - none of it from DeepSeek. Hundreds of community plugins already exist. That kind of ecosystem velocity usually takes months. Alongside the harness, DeepSeek also shipped V4 Pro, a model priced well below Opus that reportedly matches GLM 5.2 on intelligence benchmarks. The harness itself is free; the model is the only thing you pay for, and you're not locked into DeepSeek's model anyway. You can run it locally or through cloud GPU providers like Lambda. It's MIT-licensed, so there are no usage restrictions to worry about. ## Questions this post answers ### What is DeepSeek Harness (dsh) and how is it different from Claude Code? DeepSeek Harness is an open-source agent harness built on the Cordis plugin framework, where every component - models, tools, skills, memory, execution loops, and permissions - is a swappable plugin. Unlike Claude Code, which bundles a fixed set of decisions into a closed product, it can be pointed at DeepSeek's API, GLM, Gemini, a local Ollama instance, Claude Sonnet, OpenAI, or Bedrock. It is MIT-licensed and free to use. _Teams choosing between open agent frameworks and closed tools like Claude Code can track these tradeoffs on daily.dev._ ### What are the four built-in modes in DeepSeek Harness and what does Creator mode do? DeepSeek Harness ships with four modes: Standard, Code, Minimal, and Creator. Creator mode lets users author new plugins through prompting rather than writing code directly, demonstrated by adding a playable dinosaur game to the desktop app just by asking, and used by others to build code review modes, fact-checking workflows, GPU monitoring dashboards, and storyboarding tools. _Developers exploring prompt-driven plugin authoring can follow how these agent modes evolve on daily.dev._ ### How does DeepSeek Harness handle agent self-modification without breaking? An 88-page paper accompanying the release explains that every change the agent makes to its own interface or behavior comes with automatic cleanup and undo instructions baked in, making modifications reversible. This addresses a known failure mode where self-modifying agent systems tend to drift into incoherence, though whether the approach holds up at scale remains untested. _Anyone building self-modifying or long-running agents can follow this design pattern's real-world results on daily.dev._ ## Similar posts on daily.dev - [What Is an Agent Harness? The Architecture Behind Claude Code, DeepSeek Harness, and Hermes Agent](https://daily.dev/posts/what-is-an-agent-harness-the-architecture-behind-claude-code-deepseek-harness-and-hermes-agent-yewiwnntj) · freeCodeCamp · 12 upvotes · 1 comments --- Tags: [#open-source](https://daily.dev/tags/open-source), [#llm](https://daily.dev/tags/llm), [#ai-agents](https://daily.dev/tags/ai-agents), [#deepseek](https://daily.dev/tags/deepseek) [View this post on daily.dev](https://daily.dev/posts/deepseek-releases-open-source-agent-harness-with-150-000-github-stars-in-days-wi2yf4gtc) ```json {"@context":"https://schema.org","@graph":[{"@type":"Organization","@id":"https://daily.dev/#organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180},"sameAs":["https://twitter.com/dailydotdev","https://github.com/dailydotdev","https://www.linkedin.com/company/daily-dev-ltd"]},{"@type":"WebSite","@id":"https://daily.dev/#website","url":"https://daily.dev","name":"daily.dev","publisher":{"@id":"https://daily.dev/#organization"},"potentialAction":{"@type":"SearchAction","target":{"@type":"EntryPoint","urlTemplate":"https://daily.dev/search?q={search_term_string}"},"query-input":"required name=search_term_string"}}]} {"@context":"https://schema.org","@type":"TechArticle","headline":"DeepSeek releases open-source agent harness with 150,000 GitHub stars in days","url":"https://daily.dev/posts/deepseek-releases-open-source-agent-harness-with-150-000-github-stars-in-days-wi2yf4gtc","mainEntityOfPage":{"@type":"WebPage","@id":"https://daily.dev/posts/deepseek-releases-open-source-agent-harness-with-150-000-github-stars-in-days-wi2yf4gtc"},"datePublished":"2026-08-26T08:49:39.827Z","dateModified":"2026-09-13T20:03:28.641Z","description":"DeepSeek released an open-source, MIT-licensed agent harness called dsh that gained 150,000 GitHub stars within days. Built on the Cordis plugin framework with...","image":"https://pbs.twimg.com/media/HQotSQPWIAA-5ZH.jpg","thumbnailUrl":"https://pbs.twimg.com/media/HQotSQPWIAA-5ZH.jpg","isAccessibleForFree":true,"articleSection":"Collections","inLanguage":"en","publisher":{"@type":"Organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180}},"author":{"@type":"Organization","name":"Collections","logo":"https://media.daily.dev/image/upload/s--fk_6ycEi--/f_auto,q_auto/v1780996001/logos/collections?_a=BAMAMiWQ0","url":"https://daily.dev/sources/collections"},"commentCount":0,"discussionUrl":"https://daily.dev/posts/deepseek-releases-open-source-agent-harness-with-150-000-github-stars-in-days-wi2yf4gtc","interactionStatistic":[{"@type":"InteractionCounter","interactionType":{"@type":"LikeAction"},"userInteractionCount":3},{"@type":"InteractionCounter","interactionType":{"@type":"CommentAction"},"userInteractionCount":0}],"keywords":"open-source,llm,ai-agents,deepseek","timeRequired":"PT3M"} {"@context":"https://schema.org","@type":"BreadcrumbList","itemListElement":[{"@type":"ListItem","position":1,"name":"Home","item":"https://daily.dev"},{"@type":"ListItem","position":2,"name":"Collections","item":"https://daily.dev/sources/collections"},{"@type":"ListItem","position":3,"name":"DeepSeek releases open-source agent harness with 150,000 GitHub stars in days"}]} {"@context":"https://schema.org","@type":"FAQPage","@id":"https://daily.dev/posts/deepseek-releases-open-source-agent-harness-with-150-000-github-stars-in-days-wi2yf4gtc#faq","mainEntity":[{"@type":"Question","name":"What is DeepSeek Harness (dsh) and how is it different from Claude Code?","acceptedAnswer":{"@type":"Answer","text":"DeepSeek Harness is an open-source agent harness built on the Cordis plugin framework, where every component - models, tools, skills, memory, execution loops, and permissions - is a swappable plugin. Unlike Claude Code, which bundles a fixed set of decisions into a closed product, it can be pointed at DeepSeek's API, GLM, Gemini, a local Ollama instance, Claude Sonnet, OpenAI, or Bedrock. It is MIT-licensed and free to use. Teams choosing between open agent frameworks and closed tools like Claude Code can track these tradeoffs on daily.dev."}},{"@type":"Question","name":"What are the four built-in modes in DeepSeek Harness and what does Creator mode do?","acceptedAnswer":{"@type":"Answer","text":"DeepSeek Harness ships with four modes: Standard, Code, Minimal, and Creator. Creator mode lets users author new plugins through prompting rather than writing code directly, demonstrated by adding a playable dinosaur game to the desktop app just by asking, and used by others to build code review modes, fact-checking workflows, GPU monitoring dashboards, and storyboarding tools. Developers exploring prompt-driven plugin authoring can follow how these agent modes evolve on daily.dev."}},{"@type":"Question","name":"How does DeepSeek Harness handle agent self-modification without breaking?","acceptedAnswer":{"@type":"Answer","text":"An 88-page paper accompanying the release explains that every change the agent makes to its own interface or behavior comes with automatic cleanup and undo instructions baked in, making modifications reversible. This addresses a known failure mode where self-modifying agent systems tend to drift into incoherence, though whether the approach holds up at scale remains untested. Anyone building self-modifying or long-running agents can follow this design pattern's real-world results on daily.dev."}}]} ```