{"slug": "relay-open-source-coding-agent-for-non-mainstream-chinese-llm-providers", "title": "Relay – open-source coding agent for non-mainstream/Chinese LLM providers", "summary": "An open-source coding agent called Relay has been released, supporting non-mainstream and Chinese LLM providers like DeepSeek, Qwen, and GLM. The Electron app offers chat and code workspaces with file editing, command execution, and human-in-the-loop permissions, and encrypts API keys locally. It is in early public beta and welcomes community contributions.", "body_md": "An open-source, dark-mode desktop coding agent — built for people who want to use\n**non-mainstream LLM providers**, not just the big three.\n\nRelay is an Electron app that puts DeepSeek, Qwen, GLM, Kimi, MiniMax, and other\nopen/Chinese models on equal footing with the usual suspects. Bring your own API key\n(or run locally via Ollama), pick a model, and work in either a **Chat** workspace or\na full **Code** workspace with real file editing, command execution, and\nhuman-in-the-loop permissions.\n\n**Relay is an early public beta (think alpha/0.1).** It already does a lot and is\nusable day to day, but it is *not* polished production software yet — you should\nexpect bugs, half-finished corners, the occasional crash, and provider quirks\n(rate limits, model-specific errors, auth gotchas across the many supported\nproviders). That's normal for software this young.\n\n**This is being shared openly precisely so we can fix those things together.** If\nyou hit an error, please [open an issue](https://github.com/LeventeNagy/relay-coding-agent/issues)\nwith what you did, which provider/model you used, and the error text or a\nscreenshot — that's the single most helpful thing you can do. Bug reports, fixes,\nand ideas are all genuinely welcome; with more eyes on it we can squash these\nissues much faster than I can alone. See [CONTRIBUTING.md](/LeventeNagy/relay-coding-agent/blob/main/CONTRIBUTING.md).\n\nKnown in-progress areas: signed installers / auto-update, broader automated test\ncoverage, and additional renderer hardening. Your API keys are **not** affected by\nany of this — they're encrypted on your own machine and never leave it (see the\nsecurity model below).\n\n📣 **Follow @** for updates,\nprogress, and release notes.\n\n*levyathan*on Twitter/X\n\n**Many providers, one router.** DeepSeek, Alibaba/Qwen (DashScope + coding/token plans), Z.AI / GLM, Moonshot / Kimi (global + China), MiniMax, SiliconFlow, Tencent, Xiaomi, OpenRouter, and Ollama Cloud — via the Mastra model router. Add a key in Settings and its models appear in the picker.**Chat workspace.** Streaming replies, reasoning (\"thinking\") controls where the model supports them, image/document attachments, and web**search**+ deep** research**(Tavily / Brave / DuckDuckGo, key optional).** Code workspace.**Projects live in real folders. The agent reads, writes, and edits files and runs commands** scoped to the project**, with permission modes (Ask / Approve / Full).** Plan mode**asks clarifying questions and proposes a plan before touching anything.** Sources**(your design docs + framework docs) are kept in the agent's context every turn so it never loses the plot, and new projects default to a**Next.js + shadcn/ui** stack the agent scaffolds and keeps current.**Plugins (MCP).** Connect Model Context Protocol servers — Notion and Linear via one-click OAuth; GitHub, Filesystem, Git, Supabase (OAuth), Convex, Postgres, and more. Plugins are chosen per conversation and filtered by workspace.**Skills.** Reusable instruction packs the agent can pull in on demand.**Parallel sessions.** Start a run in one session and switch away — it keeps streaming and saving in the background, with a \"working\" indicator.**Context management.** Per-model context windows, automatic compaction of older turns, and a live usage meter.\n\n- API keys, plugin credentials, OAuth tokens, and search keys are encrypted at rest\nwith Electron\n`safeStorage`\n\nin the app's userData directory. - Decrypted secrets never leave the main process — the renderer only ever sees key\n*names*and connection status. - File and command tools are confined to the active project folder; writes and shell commands are gated by the permission mode (path traversal is blocked unless you grant Full access).\n- The renderer runs with\n`contextIsolation`\n\nand no Node integration; links open in your real browser, never an in-app window.\n\nSee [SECURITY.md](/LeventeNagy/relay-coding-agent/blob/main/SECURITY.md) for the full threat model and how to report issues.\n\nPrerequisites: **Node.js ≥ 20** and **Corepack** (bundled with Node). Local (stdio) MCP\nplugins like Filesystem/Git also need `npx`\n\non your PATH.\n\n```\ncorepack pnpm install     # install dependencies (pnpm 11)\ncorepack pnpm dev         # run the app in development\ncorepack pnpm build       # typecheck + production build into out/\n```\n\nThen open **Settings**, add an API key for any supported provider, and pick a model.\n\nElectron · electron-vite (Vite) · React 19 · TypeScript · [Mastra](https://mastra.ai)\n(`@mastra/core`\n\nmodel router + `@mastra/mcp`\n\n).\n\nIssues and PRs are welcome — see [CONTRIBUTING.md](/LeventeNagy/relay-coding-agent/blob/main/CONTRIBUTING.md). Run\n`corepack pnpm typecheck && corepack pnpm lint && corepack pnpm test`\n\nbefore opening a\nPR.\n\n[MIT](/LeventeNagy/relay-coding-agent/blob/main/LICENSE) © Levente Nagy", "url": "https://wpnews.pro/news/relay-open-source-coding-agent-for-non-mainstream-chinese-llm-providers", "canonical_source": "https://github.com/LeventeNagy/relay-coding-agent", "published_at": "2026-06-29 17:00:26+00:00", "updated_at": "2026-06-29 17:20:18.316872+00:00", "lang": "en", "topics": ["ai-tools", "developer-tools", "large-language-models", "ai-agents"], "entities": ["Relay", "DeepSeek", "Qwen", "GLM", "Kimi", "MiniMax", "Mastra", "Ollama"], "alternates": {"html": "https://wpnews.pro/news/relay-open-source-coding-agent-for-non-mainstream-chinese-llm-providers", "markdown": "https://wpnews.pro/news/relay-open-source-coding-agent-for-non-mainstream-chinese-llm-providers.md", "text": "https://wpnews.pro/news/relay-open-source-coding-agent-for-non-mainstream-chinese-llm-providers.txt", "jsonld": "https://wpnews.pro/news/relay-open-source-coding-agent-for-non-mainstream-chinese-llm-providers.jsonld"}}