reddit · SideProject · 10-17 12:50 52 分 ← 返回列表

I've built a harness in the style of Instinct, Muse, Dots as a side project and just open sourced it

Hi Reddit, TL;DR: While looking for a harness to use at my company, I built the engine that I'm open-sourcing today under AGPL-3.0. You can create AI assistants with long-term memory, many connectors (Google, Slack, Notion, etc) and a sandbox to run commands, all on your own server and with the models you choose. A single installation can serve a family, a company or a community, with each user isolated in their own account. The engine also works for assistants that serve the pub

五维评分

付费意愿

需求真实度

个人可落地

竞争空间

新鲜度

AI 分析

痛点：企业团队和客户需要可自托管、用户相互隔离、带长期记忆与多连接器的 AI 助手框架，现有开源框架（LangChain/Agno）在安全可控和落地体验上不满足需求

目标用户：希望安全地在内部和面向客户使用 AI agent 的中小企业、团队及技术负责人

证据：While looking for a harness to use at my company, I built the engine that I'm open-sourcing today... I started looking for a harness that both our clients and our team could use safely. After experimenting with frameworks like Agno and LangChain... I decided to build a proprietary solution from scratch.

变现：开源核心 + 托管版 SaaS 订阅、企业私有化部署与支持服务、垂直行业（如美妆零售）定制 harness，或对连接器/审计合规插件收费

MVP：聚焦单一垂直场景，做一键托管部署的 AI 助手：预置 3-5 个常用连接器 + 简单长期记忆 + 多用户隔离，免费自助试用、付费托管与合规审计增值

竞品：Agno、LangChain、Instinct、Muse、Dots、Open Claw

摘要：一位美妆科技公司的技术创始人为让公司团队与客户安全使用 AI agent，从零自建可自托管、每用户隔离、带长期记忆和 Google/Slack/Notion 等连接器的 harness，并以 AGPL-3.0 开源。他此前尝试过 Agno、LangChain 及多种 Claw 方案仍不满意，侧面验证了企业级安全可控 agent 框架的真实需求，但也说明该赛道竞品密集。