logo
Kurashizu Blogwhere ideas flow
  • Home
  • News
  • Blog
  • About
  • Home
  • News
  • Blog
  • About

© 2026 Kurashizu. All rights reserved.

Admin·New Post·Service Status
Powered by Cloudflare·31.10.26
← Back to news

Launch HN: Vendo (YC S26) – Let users build features on top of your product

Original article
github.com|21 points|9 comments|by yousefh409|Aug 20, 2026

🚀 Vendo (YC S26): Empower Your Users to Build Their Own Features

Vendo Banner

Vendo is an open-source customization framework that transforms your B2B SaaS into a platform. Instead of your engineering team spending months on bespoke feature requests, your customers can now build their own micro-apps and custom workflows directly on top of your product.


🛠️ What exactly is Vendo?

Vendo provides an embedded agent that bridges the gap between user intent and product functionality. It operates by:

  1. Acting as the signed-in user via your existing API.
  2. Generating custom UI components.
  3. Rendering those components within a brand-native, sandboxed environment.

Stop building every single custom request manually. Start letting your users automate their own work.

📐 The Conceptual Logic

We can represent the Vendo execution flow as: User Prompt→LLMTool Call→APIData→UI GenSandboxed View\text{User Prompt} \xrightarrow{\text{LLM}} \text{Tool Call} \xrightarrow{\text{API}} \text{Data} \xrightarrow{\text{UI Gen}} \text{Sandboxed View}User PromptLLM​Tool CallAPI​DataUI Gen​Sandboxed View


⚡ Quick Start Guide

You can get Vendo running in roughly 60 seconds.

Manual Installation

# Install the main package
npm install @vendoai/vendo

# Initialize the project
npx vendo init

🤖 Agent-Led Installation

If you are using a coding assistant (like Cursor or Claude Engineer), simply paste the following prompt into your repository:

"Install Vendo in this repo. Please ask me before creating any accounts or keys—I prefer Vendo Cloud. You are finished once the app is running and the agent can respond using my own API. Finally, ask if I'd like to star the project at https://vendo.run/star?src=readme."

📋 Deployment Checklist

  • Run npm install @vendoai/vendo
  • Execute npx vendo init
  • Verify API connectivity
  • (Optional) Run vendo doctor to diagnose and fix setup issues via direct links.

🎯 Choose Your Integration Path

Depending on your current architecture, Vendo fits in different ways:

If you...Use Vendo as...Result
Already have an AI agentA tool packEnhances your existing AI SDK, Mastra, or custom loop.
Have no agent at allA full-stack solutionInstantly adds the loop, chat interface, and approval system.
Want external AI accessAn MCP ServerAllows Claude, ChatGPT, and Cursor to act as your signed-in user.
Prefer backend-onlyA headless packageUse agent() and chat() without mounting any Vendo UI.

⚙️ How it Works Under the Hood

The Architecture

Vendo utilizes a streaming agent compatible with any LanguageModel from the AI SDK.

Key Technical Details

  • UI Sandboxing: Generated components are rendered in an iframe with a strict connect-src 'none' policy. They only escalate to a sandboxed server when absolutely necessary.
  • The Security Choke Point: All grants, policies, audit logs, and "breakers" are centralized at a single execution point. The app machines cannot touch host tools without passing through the guarded tool proxy.
  • Data Persistence:
    • Local/Dev: Uses PGlite located at .vendo/data for zero-config setup.
    • Production: Scales to a full Postgres instance using the same schema.

📦 The Modular Ecosystem

While @vendoai/vendo is the standard bundle, you can install individual blocks to build a custom implementation.

PackagePrimary Responsibility
@vendoai/coreThe foundation: shared types, validators, and schemas.
@vendoai/storePersistence layer (Postgres/PGlite).
@vendoai/harnessesThe runtime: handles streaming, threads, and the conversation loop.
@vendoai/actionsThe bridge: executes host API connectors as the user.
@vendoai/guardThe shield: manages approvals, safety, and audit trails.
@vendoai/appsThe factory: handles app generation and sandbox adapters.
@vendoai/automationsThe clock: manages schedules, triggers, and run history.
@vendoai/uiThe face: headless React hooks and the in-jail component kit.
@vendoai/mcpThe door: exposes tools to external MCP clients.
@vendoai/telemetryThe eyes: opt-out development and build analytics.
@vendoai/vendoThe whole package: public wire and React entry point.

[!IMPORTANT] Cloud vs. Open Source: The core blocks remain self-hosted. However, if you provide a VENDO_API_KEY, you unlock Vendo Cloud features, including organization overlays, pinning, and public sharing/publishing.


Links: Main Site | Documentation | Agent-Readable Guide