GumloopGumloop
Gumloop 是画布+连接器形态的企业向 AI agent 平台，多模型选择器是显性卖点（首页当日招牌：Build AI agents for work；Agent Settings 里 Model 可选「Smartest (Claude 5.5 Opus)」，用量面板按 GPT-6.1 Sol / Claude Opus 5.5 / Grok 4.7 / Kimi K3 / DeepSeek V4.1 Flash 归因 credits 消耗）：GTM/客服场景用例为主轴，法律主体是 AgentHub Inc.。
Gumloop is a canvas+connectors enterprise AI agent platform with the multi-model picker as an explicit selling point (same-day homepage headline: Build AI agents for work; Agent Settings' Model item offers "Smartest (Claude 5.5 Opus)", the usage panel attributes credits across GPT-6.1 Sol / Claude Opus 5.5 / Grok 4.7 / Kimi K3 / DeepSeek V4.1 Flash): GTM/support use cases as the main axis; the legal entity is AgentHub Inc.
适合与不适合When to use it, when not to
适合：GTM/客服多模型 agent（画布+连接器+Skills）、要按 credits 统一治理各家模型用量。 不适合：要自托管（社区复刻≠官方）、要大陆直连备案入口、要免费无限量（credits 制）。
Fits: multi-model agents for GTM/support (canvas + connectors + Skills), unified credits-based governance of per-vendor model usage. Not a fit: self-hosting (a community clone ≠ official), a filed direct-mainland entry, free unlimited usage (credits metering).
中文生态China availability 国产 / 国内可直接用 / 备案状态 · 快照domestic / direct access / CAC filing · snapshot 2026-10-30
- 出品属地Origin
- 国外出品foreign-made AgentHub Inc. (doing business as Gumloop)（ToS 当日直抓在册）。AgentHub Inc. (doing business as Gumloop), on record in the ToS direct-fetched same day.
- 国内使用Direct access
- 大陆可直连direct OK www.gumloop.com、/tos、/privacy-policy、/changelog 大陆直连实测 200（2026-10-30）。www.gumloop.com, /tos, /privacy-policy and /changelog measured 200 direct from mainland (2026-10-30).
- 备案状态CAC filing
- 未见于备案名单not in filing list：未见于网信办公示名单（2026-10-30 检索百炼公示页与第十六批批次清单，未见 Gumloop/AgentHub 条目）；该产品未面向大陆公众提供服务，备案不适用。not in the CAC public registry (Bailian disclosure page and the 16th batch list searched 2026-10-30; no Gumloop/AgentHub entry); the product does not serve the mainland public, so filing does not apply.
备案信息以网信办《生成式人工智能服务已备案信息》公示名单为准（按批次滚动更新，本站只抄编号不解读）；「大陆可直连」是本站从大陆网络直连实测的结果，不是厂商承诺。GitHub 类站点当日实测有波动，口径写在依据里。Filing status is quoted from the CAC public registry (updated batch by batch; this site copies the number and does not interpret it). "Direct access" is this site's own reachability measurement from a mainland network, not a vendor promise. GitHub-hosted sites fluctuated on the measurement day — the note says so.
固定坐标系Fixed coordinate system
8 个维度dimensions, 与同分区其他对象逐项可比each comparable item by item against others in the same section.
- 模型与开放条件Model & openness
多模型选择器是产品显性面（首页当日在册：Agent Settings 的 Model 项「Smartest (Claude 5.5 Opus)」； 用量面板列 GPT-6.1 Sol / Claude Opus 5.5 / Grok 4.7 / Kimi K3 / DeepSeek V4.1 Flash 各家消耗）—— 模型按 credits 消耗计费，各家可用清单以界面为准。
The multi-model picker is an explicit product surface (same-day homepage: the Model item in Agent Settings shows "Smartest (Claude 5.5 Opus)"; the usage panel lists per-vendor consumption for GPT-6.1 Sol / Claude Opus 5.5 / Grok 4.7 / Kimi K3 / DeepSeek V4.1 Flash) — models are billed via credits; the available list per vendor follows the UI.
- 运行位置Where it runs
云托管形态（www.gumloop.com 直连 200）；未见官方自托管发行声明（社区有 open-gumloop 复刻仓， 非官方）；agent 运行在 Gumloop 云端，按 credits 计量。
Cloud-hosted form (www.gumloop.com direct 200); no official self-hosted distribution found (a community clone repo open-gumloop exists, unofficial); agents run in Gumloop's cloud, metered by credits.
- 本地文件Local files
工作流与 agent 数据在 Gumloop 云端；本地文件概念不适用——产物经连接器（Slack/HubSpot/Gong/Gmail/ Salesforce）落在第三方系统；SDK（gumloop-py/js）提供代码侧接入面但运行时仍在云端。
Workflow and agent data live in Gumloop's cloud; the local-file concept does not apply — artifacts land in third-party systems via connectors (Slack/HubSpot/Gong/Gmail/Salesforce); the SDKs (gumloop-py/js) offer a code-side entry but the runtime stays in the cloud.
- 关机后的任务Tasks after shutdown
agent 常驻与定时运行在册（首页当日用例面板：Renewal Watch / Ticket Triage 等值守型 agent）； 排队与失败重试细节未逐条核验。
Always-on and scheduled agents are on record (same-day homepage use-case panel: Renewal Watch / Ticket Triage watch-type agents); queueing and retry details not verified item by item.
- 工具与扩展Tools & extensions
Connectors 生态当日在册（Slack/HubSpot/Gong/Gmail/Salesforce，View all 入口）；Skills 面在册 （Account Research/Outbound Email/Call Summary/CRM Hygiene）；guMCP_template 仓在册（MCP 方向官方模板）。
The Connectors ecosystem is on record same day (Slack/HubSpot/Gong/Gmail/Salesforce, with a View all entry); the Skills surface is on record (Account Research/Outbound Email/Call Summary/CRM Hygiene); the guMCP_template repo is on record (an official MCP-direction template).
- 上下文与记忆Context & memory
agent 指令（Add instructions）与 Skills 是持久配置面；跨会话记忆机制未核验到官方声明，按未知处理—— 知识面主要靠连接器拉取的外部数据（CRM/Gong 记录）。
Agent instructions (Add instructions) and Skills are the persistent configuration surface; cross-session memory is not verified in official statements, treated as unknown — the knowledge surface mainly comes from connector-pulled external data (CRM/Gong records).
- 权限与限制Permissions & limits
企业向（导航 Enterprise 栏目在册）；ToS 主体 AgentHub Inc. dba Gumloop（当日在册）； credits 计费模式在册；训练用途开关路径未逐条核验。
Enterprise-facing (an Enterprise nav column is on record); ToS entity AgentHub Inc. (doing business as Gumloop) (on record same day); credits billing on record; training-use toggle path not verified item by item.
- 适合什么任务Best for
适合：GTM/客服场景的多模型 agent（画布+连接器）、要按 credits 统一用量治理（Model Providers 面板）。 不适合：要自托管、要大陆直连备案入口、要免费无限量（credits 制）。
Fits: multi-model agents for GTM/support (canvas + connectors), unified credits-based usage governance (the Model Providers panel). Not a fit: self-hosting, a filed direct-mainland entry, free unlimited usage (credits metering).
头号误解Top misconception
- 条款主体是 AgentHub Inc. (doing business as Gumloop)——品牌名与法律主体不同名，引用主体按 ToS 原文The ToS entity is AgentHub Inc. (doing business as Gumloop) — brand and legal entity differ; cite the entity as written in the ToS
- 条款路径是 /tos（/terms、/legal/terms 均 404）——按惯例路径探测会误判「无条款页」The terms path is /tos (/terms and /legal/terms both 404) — probing conventional paths will misjudge "no terms page"
- 多模型面板显示的是用量归因（各家 token 消耗）不是「可任意指定任意模型」——具体可选清单以界面为准The multi-model panel shows usage attribution (per-vendor token consumption), not "any model can be freely specified" — the selectable list follows the UI
未知项清单Known unknowns
- 「Smartest」模型路由的判定逻辑（未核验到官方说明）
- 自托管/私有部署形态（未核验到官方声明；open-gumloop 是社区复刻）
- 跨会话记忆机制（未核验到官方声明）
- 训练用途开关路径与默认态（未逐条核验）
- The decision logic of "Smartest" model routing (no official explanation verified)
- Self-hosting/private deployment form (no official statement verified; open-gumloop is a community clone)
- Cross-session memory mechanism (no official statement verified)
- Training-use toggle path and default state (not verified item by item)
证据来源Evidence sources
判断可回到以下一手源复核。本站核验日Every judgement can be rechecked against the primary sources below. Verified on 2026-10-30，内容更新日content updated on 2026-10-30。
| 类型Type | 名称Name | 链接Link |
|---|---|---|
| official | Gumloop · 官网（Build AI agents for work、Model/Connectors/Skills 面板，2026-10-30 直抓 200）Gumloop · official site (Build AI agents for work; Model/Connectors/Skills panels, direct-fetched 200 on 2026-10-30) | https://www.gumloop.com |
| official | Gumloop · Terms of Service（主体 AgentHub Inc. dba Gumloop，更新 2026-10-07，2026-10-30 直抓 200）Gumloop · Terms of Service (entity AgentHub Inc. dba Gumloop, updated 2026-10-07, direct-fetched 200 on 2026-10-30) | https://www.gumloop.com/tos |
| official | Gumloop · Privacy Policy（2026-10-30 直抓 200）Gumloop · Privacy Policy (direct-fetched 200 on 2026-10-30) | https://www.gumloop.com/privacy-policy |
| changelog | Gumloop · Changelog（最新条目 2026-10-28，2026-10-30 直抓 200）Gumloop · Changelog (latest entry 2026-10-28, direct-fetched 200 on 2026-10-30) | https://www.gumloop.com/changelog |
| repo | gumloop/gumloop-py · 官方 Python SDK（Apache-2.0，GitHub API 当日核验）gumloop/gumloop-py · official Python SDK (Apache-2.0, GitHub API verified same day) | https://github.com/gumloop/gumloop-py |
实测记录Measured runs
本档未跑工作流搭建实测；结论只覆盖平台能力与权限边界。
实测建议（三条，都指向官方可读页没给的东西）：
- 各家模型的实时可选清单与路由规则（「Smartest」自动路由的判定逻辑未在册）
- credits 单价与各家模型的消耗率（面板显示消耗量，换算率未在册）
- 连接器授权的数据权限边界（Salesforce/Gong 级连接的读写面未逐条核验）
This site has not completed measured runs. The coding-oriented test protocol is at protocols/coding/_protocol.md (the coding task suite does not apply to agent platforms).
Suggested measurements (three, all pointing at things the official readable pages do not give):
- The live per-vendor model list and routing rules (the decision logic behind "Smartest" auto-routing is not on record)
- Credit pricing and per-vendor consumption rates (the panel shows consumption volume; the conversion rate is not on record)
- Data permission boundaries of connector authorizations (read/write scopes of Salesforce/Gong-level connections not verified item by item)
相关条目Related entries
- LindyLindy同为工作 agent 平台：Lindy 是雇佣式「AI 员工」，Gumloop 是画布+连接器的自建 agent。Also a work-agent platform: Lindy is the hired "AI employee", Gumloop the self-built canvas+connectors agent.
- n8nn8n集成自动化对照：n8n fair-code 可自托管；Gumloop 全托管、agent 与多模型治理为主轴。Integration-automation contrast: n8n is fair-code and self-hostable; Gumloop is fully hosted with agents and multi-model governance as the main axis.
- LangChainLangChain代码优先对照：要源码级控制归 framework，要托管画布归本类。The code-first contrast: source-level control belongs to the framework class, hosted canvas to this class.
本页由This page is generated from the ai-handbook 数据层生成（CC BY 4.0）。data layer (CC BY 4.0).
方法论与坐标系定义见仓库内For the methodology and coordinate definitions, see METHODOLOGY.md.