archestack/best-of-ai-starters：89 个开源 AI 应用起步模板排名榜

AI starters, ranked. 🏆

Templates you fork to ship your own AI product: what is wired in, what is missing, no fluff. Paid never moves a rank.

Best of AI Starters is a ranked list of 89 open-source starter templates for building AI apps: chat apps, RAG, agent backends, agent UIs, voice, MCP servers, SaaS boilerplates and more, most of them Next.js, Python, LangGraph or the AI SDK. Each entry says what is wired in, what you still have to build, and carries a score from public GitHub signals. _{Updated 2026-10-09 (UTC).}

Tip

🔎 Every template fact-checked on GitHub before it is listed  ·  ✍️ Every entry says what is wired in and what you still have to build  ·  ▶️ Demo links so you can see it running first

Looking for finished AI apps you install and use? 👉 Best of Self-Hosted AI

🔥 Top 10 right now

_{89 projects ranked in 12 categories · data from 2026-10-09 · re-ranked twice a day}

_{How to read it: rank 🥇🥈🥉 then #⁠4, #⁠5… (equal scores: more stars first) · score out of 100 sets the order: hover it to see what it is made of, or read how we rank · click a project name for its review · ↗ opens another site (Cmd/Ctrl-click for a new tab)}

_{Ordered by score: adoption, freshness, how easy it is to run, and a few more signals as we measure them (how we rank).}

⚡ Pick by vibe

• 💬 Ship a chat app this weekend → Chat apps
• 📚 RAG over my data, with citations → RAG and search
• 🧩 An agent backend I can extend → Agent backends
• 🖥️ A UI that shows what the agent is doing → Agent UI and generative UI
• 🎙️ A voice agent that talks back → Voice and realtime
• 🔌 Build an MCP server → MCP servers and chat-host apps
• 💳 SaaS with auth, billing and AI built in → SaaS boilerplates with AI
• 🏗️ A base for my own app builder or coding agent → App builders and coding agents
• ✏️ An AI editor or workflow canvas → AI editors and workflow canvases
• ☁️ A reference architecture for my cloud → Cloud reference architectures
• ⚙️ Just the API backend → AI API backends
• 📱 Mobile app or browser extension → Mobile and browser extensions

🗂️ All categories

• 💬 Chat apps · 12
• 📚 RAG and search · 12
• 🧩 Agent backends · 16
• 🖥️ Agent UI and generative UI · 6
• 🎙️ Voice and realtime · 13
• 🔌 MCP servers and chat-host apps · 5
• 💳 SaaS boilerplates with AI · 5
• 🏗️ App builders and coding agents · 5
• ✏️ AI editors and workflow canvases · 4
• ☁️ Cloud reference architectures · 3
• ⚙️ AI API backends · 6
• 📱 Mobile and browser extensions · 2

💬 Chat apps

Chat interfaces and single-feature text apps you fork as the base of a conversational product. _{12 ranked · 📝 all reviews}

langchain-nextjs-template leads with 67, ahead of chatbot (58) and claude-quickstarts (53).

💡 How to choose

• Pick by framework first (Next.js, Nuxt, Laravel); porting a chat UI across frameworks costs more than any feature gap.
• Check what is persisted (chat history, files, users) and which database it assumes before you commit.
• Prefer starters that route models through one provider layer so you can swap vendors later.

↑ categories

📚 RAG and search

Retrieval over your own documents or data, answer engines, and natural-language-to-SQL starters. _{12 ranked · 📝 all reviews}

morphic leads with 75, ahead of azure-search-openai-demo (56) and rag-postgres-openai-python (53).

💡 How to choose

• Match the vector store to what you already run (pgvector, Azure AI Search, a hosted index).
• Look at how ingestion works; a good chat UI with no re-indexing story will stall in production.
• Answer engines need search API keys; count those costs before picking one.

↑ categories

🧩 Agent backends

Agent templates and scaffolds (LangGraph, ADK, OpenAI Agents SDK, Cloudflare Agents, eve) meant to be extended. _{16 ranked · 📝 all reviews}

eve-software-factory-template leads with 67, ahead of marketing-team-eve-template (61) and personal-agent-template (59).

💡 How to choose

• Choose the framework you want to live with; templates are thin and the framework is the real dependency.
• Check where state lives (Durable Objects, Postgres, a managed platform) and whether that fits your hosting.
• Favor templates that ship tests or an eval hook so agent changes can be checked.

↑ categories

🖥️ Agent UI and generative UI

Frontends that render agent steps, tool calls, approvals or model-generated components. _{6 ranked · 📝 all reviews}

agent-chat-ui leads with 49, ahead of agent-ui (45) and OpenGenerativeUI (45).

💡 How to choose

• Confirm the UI speaks your backend's protocol (LangGraph SDK, AG-UI, AI SDK streams).
• Check human-in-the-loop support if your agent needs approvals.
• Generative UI that renders model HTML needs sandboxing; prefer starters that isolate it.

↑ categories

🎙️ Voice and realtime

Voice agents, realtime speech-to-speech apps and their web, phone and native clients. _{13 ranked · 📝 all reviews}

agent-starter-react leads with 54, ahead of agent-starter-python (52) and agent-starter-node (52).

💡 How to choose

• Pick the transport first (LiveKit, Pipecat/Daily, provider websockets); clients and agents must match.
• Check turn detection and interruption handling; it decides how natural the agent feels.
• Separate frontend and agent starters usually need to be combined; plan for both.

↑ categories

🔌 MCP servers and chat-host apps

Templates for building MCP servers and apps that run inside chat hosts such as ChatGPT. _{5 ranked · 📝 all reviews}

mcp-typescript-template leads with 60, ahead of template-mcp-server (47) and mcp-for-next.js (42).

💡 How to choose

• Decide stdio vs remote HTTP early; remote servers need auth (OAuth) from day one.
• Pick the language of the system you are exposing, not the language of the client.
• Check the protocol version the template targets; transports have changed more than once.

↑ categories

💳 SaaS boilerplates with AI

Product boilerplates with auth, billing and data that already include AI features or agent access. _{5 ranked · 📝 all reviews}

velobase-harness leads with 60, ahead of open-saas (55) and AI-Fullstack-SaaS-Boilerplate (41).

💡 How to choose

• Verify the AI part is real code (model calls, usage metering), not only editor config files.
• Usage-based billing matters for AI costs; check whether metering is included or left to you.
• Weigh the framework lock-in (Wasp, tRPC, Django) against what your team knows.

↑ categories

🏗️ App builders and coding agents

Prompt-to-app builders and platforms that run coding agents in sandboxes. _{5 ranked · 📝 all reviews}

llamacoder leads with 56, ahead of open-agents (51) and vibesdk (49).

💡 How to choose

• Check which sandbox runs generated code (E2B, Vercel Sandbox, Cloudflare) and its pricing.
• Look at how repos and credentials are connected; these apps act on real code.
• Expect to bring several API keys; the templates are platforms, not single-page demos.

↑ categories

✏️ AI editors and workflow canvases

Rich-text editors with AI commands and node-based canvases for chaining model calls. _{4 ranked · 📝 all reviews}

plate-playground-template leads with 54, ahead of editor (51) and workflow-builder-template (39).

💡 How to choose

• For editors, pick the editor engine you can extend (TipTap, Plate) before the AI features.
• For canvases, check whether workflows persist server-side or only in the browser.
• Collaboration features add infrastructure; skip them if you do not need multi-user editing.

↑ categories

☁️ Cloud reference architectures

Vendor reference apps and infrastructure-as-code for running AI apps on Azure or Google Cloud. _{3 ranked · 📝 all reviews}

agent-landing-zone leads with 63, ahead of azurechat (44) and openai-chat-app-quickstart (38).

💡 How to choose

• Only pick these if you are already on that cloud; the value is in the infra wiring.
• Check the IaC tool (azd/Bicep, Terraform) and the identity model before forking.
• Budget for managed services these templates provision by default.

↑ categories

⚙️ AI API backends

Backend service templates (FastAPI, Express, Hono) that expose models or agents over an API. _{6 ranked · 📝 all reviews}

agent-service-toolkit leads with 73, ahead of fastapi-langgraph-agent-production-ready-template (59) and full-stack-ai-agent-template (59).

💡 How to choose

• Check auth, rate limiting and tracing; these are what separate a template from a demo.
• Prefer templates with Docker and tests if the service will run in production.
• Make sure the model layer is swappable (LiteLLM, provider adapters) if you expect to change vendors.

↑ categories

📱 Mobile and browser extensions

Native, cross-platform mobile and browser-extension starters with AI features built in. _{2 ranked · 📝 all reviews}

react-native-ai leads with 53, ahead of extro (39).

💡 How to choose

• Keep API keys off the device; prefer starters with a server proxy.
• Check platform coverage (iOS, Android, web, Chrome/Firefox) against your targets.
• On-device or browser built-in models change quickly; verify the APIs used are still current.

↑ categories

🏅 How we rank

Every project gets a score out of 100 from the signals below. A signal we cannot measure yet is left out and its weight goes to the others, so nobody loses points for things we have not checked. The score sets the order; 🥇🥈🥉 mark the top three places in each table.

_{Penalties: no license −10 · data must leave your host −10 · star-growth anomaly −15 · unpatched security advisory −20. Archived projects are removed. Paid placements never touch the score.}

_{The weights are our judgment, not science. Think one is wrong? Tell us. Method changes: 2026-10-09: Docker alone earns no points, medals mark places 1–3 instead of score bands.}

🧠 How this works

• 🧩 Starter here means a template or reference app you clone and build on, not a finished app or a library.
• 🔎 Found in curated lists, app stores and template galleries, then fact-checked on GitHub: stars, last commit, license, Docker files.
• ✍️ Written from the README, never copied from other lists: what it does, what it needs, strengths and weaknesses as claims you can check. Specs say unknown rather than guess.
• 🔄 Kept fresh: entries are rewritten when the README or the latest release changes; projects quiet for 12 months get marked stale, archived ones drop off.
• 🚫 No pay-to-rank. Sponsors and affiliate links, if any, are labeled and never touch the order.

📬 Your turn

➕ Add a project · 🛠️ Fix wrong data · 🚪 Opt out (done within 24 hours). Everything here is generated, so use the forms instead of editing files.

📚 Where candidates come from

These lists, app stores and galleries (facts and links only, no text copied), plus community submissions: awesome-llm-apps · awesome-saas-boilerplates · awesome-opensource-boilerplates · awesome-langgraph · awesome-langchain · awesome-supabase · voiceai · awesome-nextjs · gh-search-ai-templates · gh-search-llm-starters · gh-search-rag-templates · vercel · langchain-ai · openai · anthropics · livekit-examples · pipecat-ai · copilotkit · assistant-ui · cloudflare · run-llama · azure-samples · aws-samples · google-gemini · googlecloudplatform · supabase-community · get-convex.

📄 License

Data (data/, this README, reviews/) is CC BY 4.0; see LICENSE-DATA. Code is MIT; see LICENSE. Project names and descriptions belong to their owners.

_{Made with ☕ by Archestack · see also Best of Self-Hosted AI}