A curated list of terminal products, the services agents call, the models that drive them, and open standards.

This repository is Awesome-Agent-Ecosystem.

🌐 Live Website: https://0xwelt.github.io/Awesome-Agent-Ecosystem/

Content structure: entries are organized in the contents/ directory — one markdown file per entry, grouped by section, category, and subcategory, with _meta.md files describing each folder. The live website is generated from these files by yaal/scripts/parse-readme.js.

One list with three super-classes, plus thin Standards. Terminal Products are what a person opens: General Chat, Coding Agents, Vibe Coding Communities, Personal Agents, and AI-enhanced Apps. Agent Services are the tools and platforms those products call, including infrastructure and official CLIs. LLMs are the models that drive agents, grouped by the company that trains them. Open protocols and formats stay in Standards. Each product is listed in one file, except a unified chat-and-agent desktop, which is also listed under Personal Agents.

Terminal Products are the five sections a person opens. General Chat is the conversation app itself. Coding agents run the agent, including Lovable-style builders that create apps in the browser. They commonly offer plan mode, goals, compact, hooks, subagents, dynamic workflows, and programmatic tool calling (PTC); those are capabilities, not entries. Vibe coding communities are where people create apps with AI and share them. Personal agents are named products that act for a person over time. They are filed flat under Personal Agents, and runtime and capability forms are category-scoped tags. AI-enhanced apps are traditional products whose main form is still that category, with the interaction rebuilt around AI. Agent services are the tools and platforms those products call. LLMs are the models behind them, not the chat apps and not inference hosts. Standards are open protocols and formats.

Cowork, coding, and computer-use are capabilities, not top-level classes. General Chat always keeps the conversation product, including ChatGPT, Claude, Gemini, Kimi, and Doubao. If the vendor has officially merged chat and agent or work into one desktop app, also list that unified desktop under Personal Agents. ChatGPT Desktop, Claude Desktop, and Gemini for desktop are that case. If the vendor still splits a general-chat line from a separately named agent or work line, such as Kimi versus Kimi Work or Doubao versus Doubao Work, Personal Agents keeps only the named agent or work product. Do not add that vendor's chat-client installer as a second file. Extra named always-on agents, such as Dots and Claude Duet, stay their own files in the same folder. Record how you reach the agent and where the work runs with that section's tags. Coding-primary products stay in Coding Agents. Local coding agents record official interface surfaces with that subclass's tags.

General Chat does not include Claude Code, Gemini CLI, GitHub Copilot, Doubao Work, QwenWork, Kimi Work, or Nano Work. An AI-enhanced app is an independent terminal. A suite that only added an AI button is not an entry there. Do not file the same product twice, except the unified desktop case above.

flowchart TB
    Person["👤 Person"]

    subgraph products["🖥 Terminal Products"]
        direction TB
        subgraph chat["💬 General Chat"]
            direction LR
            GC1["Conversation apps"]
        end
        subgraph coding["🛠 Coding Agents"]
            direction LR
            H1["Local Coding Agents"]
            H2["Standalone IDEs"]
            H3["IDE Extensions"]
            H4["Web builders"]
            H5["Cloud agents"]
            H6["Mobile"]
            H7["Agent SDKs"]
        end
        subgraph community["✨ Vibe Coding Communities"]
            direction LR
            C1["AI app builders"]
        end
        subgraph personal["🤖 Personal Agents"]
            direction LR
            P1["Named agents"]
        end
        subgraph apps["🧩 AI-enhanced Apps"]
            direction LR
            A1["AI browsers"]
        end
    end

    subgraph services["🧰 Agent Services"]
        direction LR
        S1["Fulfillment"]
        S2["Infrastructure"]
        S3["Orchestration and tools"]
    end

    subgraph llms["🧠 LLMs"]
        direction LR
        L1["OpenAI"]
        L2["Anthropic"]
        L3["Google"]
        L4["xAI"]
        L5["DeepSeek"]
        L6["Alibaba"]
        L7["Moonshot"]
        L8["Zhipu"]
        L9["Meta"]
        L10["Mistral"]
    end

    subgraph standards["📐 Standards"]
        direction LR
        ST1["Protocols"]
        ST2["Formats"]
    end

    Person --> chat
    Person --> coding
    Person --> community
    Person --> personal
    Person --> apps
    chat -->|calls| services
    coding -->|calls| services
    personal -->|calls| services
    apps -->|calls| services
    coding -.->|driven by| llms
    coding -.->|follows| standards

Named personal agents that act over time live directly in contents/products/personal-agents/. Runtime and capability forms are category-scoped tags on the entry, not exclusive subfolders.

Interaction surfaces are desktop (official desktop or on-device app), web (official web app), and mobile (official mobile app). remote means a phone or a chat channel can dispatch or control work on a connected computer. Execution locus is local (the work runs on a computer the user provides, so scheduled or overnight work typically needs that machine awake), cloud (the work runs on vendor infrastructure and can continue after a local client is closed), or self-hosted (the user runs the agent runtime or gateway). Capabilities the entry already supports stay as computer-use, skills, routines, and connectors. These tags combine freely. skills on a personal agent means the product supports agent skills, which is separate from the Agent Services connection tag skill. Do not add industry or vertical tags.

General Chat always keeps the conversation product, including ChatGPT, Claude, Gemini, Kimi, Doubao, and the other chats in General Chat. That set is not a closed list. If the vendor has officially merged chat and agent or work into one desktop app, that unified desktop is also listed here. ChatGPT Desktop, Claude Desktop, and Gemini for desktop are that case. If the vendor still splits a general-chat line from a separately named agent or work line, such as Kimi versus Kimi Work or Doubao versus Doubao Work, this folder keeps only the named agent or work product. Do not add that vendor's chat-client installer as a second file. Extra named always-on agents, such as Dots and Claude Duet, stay their own files in the same folder.

Local or self-hosted coding agents live in contents/products/coding-agents/local-coding-agents/. They typically include a terminal, and may also have a web UI or desktop app. Official interface surfaces are category-scoped tags on that entry, and they combine freely.

cli is an official terminal, TUI, or CLI. web is an official web UI, whether it is served locally or hosted. desktop is an official standalone desktop app. ide is an official IDE extension or first-party IDE companion, including an official ACP or IDE integration. Unofficial third-party plugins are not ide. mobile is an official mobile app, including a public TestFlight when it is official. Do not add tags outside that set. One product line stays one file. Do not copy it into IDE Extensions or Mobile Harnesses because it also has those surfaces. Standalone IDEs, IDE Extensions, web builders, cloud agents, mobile harnesses, and agent SDKs stay sibling folders and do not use tags.

Entries are the tool or service. Connection methods are tags, not categories. The tag vocabulary is mcp (official or vendor-hosted MCP server), skill (an Agent Skill the vendor ships), grok-bot (listed in the Grok Bot Connect Apps catalog), cli (official command-line interface), and chatgpt (official ChatGPT app or Actions published by the vendor). Add a tag only when that surface is real. Keep the word skill here.

Tags are not a global vocabulary. Agent Services tags are connection methods only. Personal Agents uses the separate capability and runtime vocabulary above. Local Coding Agents uses the interface vocabulary above. The Agent Services tag cli is a connection method, and it is a different category scope from the Local Coding Agents tag cli. General Chat, the other Coding Agents subclasses, Vibe Coding Communities, AI-enhanced Apps, LLMs, and Standards do not use tags.

Create a markdown file in the matching contents/ folder:

---
name: Tool Name
link: https://example.com
---

Short description of the entry.

Required frontmatter is name and link. Leave out a command field. Use the canonical product URL, without UTM, referral, or affiliate parameters. An optional image must be an absolute https URL of the product's official preview. Do not invent one.

Put a conversation app in General Chat. Put a coding agent or Lovable-style web builder in Coding Agents. Put a local or self-hosted coding agent in Local Coding Agents and record official interface surfaces with that subclass's tags. Put a vibe coding community in Vibe Coding Communities. Put a named personal agent directly in Personal Agents, and record runtime and capability forms with that section's tags. Put an independent terminal whose main form is still a traditional category, such as an AI browser, in AI-enhanced Apps. Put a model under LLMs in the company folder of the lab that trains it. Put an open protocol or format in Standards. Put a supporting tool or service in Agent Services, directly in one of the 17 category folders. The same product must not appear as two files, except the unified desktop case above: the conversation product stays in General Chat, and the merged desktop app is also listed under Personal Agents.

The entry is picked up when the website builds. Prefer adding a separate file over editing an existing one — it keeps PRs small and conflict-free.

This project exists thanks to all the people who contribute.

This work is licensed under a Creative Commons Attribution 4.0 International License.