Vendo

Vendo is an open-source AI customization and embedded-agent platform designed for software companies that want their customers to build their own features and micro-apps on top of an existing product.

Rather than functioning as a standalone project-management, CRM, or productivity application, Vendo operates as a layer embedded into another company's software. Customers can describe what they need in natural language, and Vendo's agent can assemble a working interface using the host product's existing APIs, components, permissions, and design system.

The central concept is the embedded agent. Vendo defines an embedded agent as an AI agent that lives inside another company's product and acts through that product's own API as the signed-in customer. Instead of merely answering questions about the product, the agent can operate the product and create working interfaces, records, automations, and views.

Vendo's generated interfaces are intended to be persistent product functionality rather than disposable AI responses. A customer can describe a requirement such as a custom dashboard, spending monitor, deadline view, workflow, or alert, and Vendo can assemble a micro-app containing UI, data, logic, actions, state, and automations.

The platform is designed to learn the host application's existing architecture. Running npx vendo init scans the application repository, identifies the framework, components, theme, APIs, and permission rules, and creates the configuration required for Vendo to operate within that product.

Vendo extracts the product's design language into theme information, maps API endpoints into callable tools, and captures permission policies. This allows generated interfaces to use the product's own visual language and operate through the same authorization boundaries as the rest of the application.

The generated UI can be embedded into different surfaces, including overlays, chat sidebars, full panels, or the host product's own components. The same customer-created application can also be accessed through supported AI environments such as Claude and ChatGPT.

Vendo supports standing automations in addition to on-demand interfaces. A customer can describe a recurring requirement such as running an operation every Monday, and Vendo can turn that request into an automation that executes on a schedule with the permissions already granted to that customer.

The platform also supports integrations with external services. Vendo's current integration catalog lists 120 applications, including categories such as team collaboration, developer tools, project management, CRM, accounting, task management, productivity, and documents. Connected services can be authorized by users and subsequently accessed by the embedded agent on their behalf.

A major differentiator is customer-specific customization without requiring the SaaS vendor to manually build every requested feature. Instead of putting every customer request into a product roadmap or support-ticket queue, the customer can potentially construct a working view or workflow themselves through the embedded agent.

Vendo also incorporates governance controls. Generated applications operate within the host product's permission model, while risky actions can require approval. Generated UI is sandboxed, and tool calls pass through a guarded execution layer. Activity is recorded in an append-only audit log.

The platform's execution model therefore attempts to preserve the host application's existing authority rather than creating a separate unrestricted AI identity. The agent acts as the signed-in customer and can only access what that customer's existing permissions allow.

Vendo is Apache-2.0 open source. The single-player agent and core customization layer can be self-hosted, while Vendo Cloud adds multiplayer and organizational capabilities such as hosted sharing, publishing, organization overlays, and pinning.