Skip to main content
If you’re building agents in code (support bots, shopping assistants, etc.), you’re really doing two completely different jobs:
  1. The AI Engineering – models, prompts, tools, RAG, memory, evals, monitoring, etc.
  2. Everything around it – session storage, APIs, session viewer, playgrounds for stakeholders, user management, email/slack integrations, etc.
The first part is fun, but the second part can take a ridiculous amount of time. The goal of AgentView is to handle (2) so you can focus on (1). AgentView is completely framework-agnostic. You write a stateless AI logic in Python or TypeScript, with any framework or none at all. The AgentView API and SDK provides a persistence layer that stores and orchestrates your session state, so you never have to build or maintain a complex backend. AgentView also includes Studio: a visual interface for exploring and managing your agents. It’s designed for ease of use by non-technical users, with built-in support for collaboration. Studio is extremly customisable, you can provide your custom components for any part of the experience.

​
Studio features

  • Browse and inspect real agent sessions
  • Invite teammates; multi-user by default
  • Comment, mention, and get notified (Notion-style)
  • Use a realistic playground: create private test sessions, share with stakeholders when ready
  • Customize deeply: define how sessions, cards, inputs, and custom pages render by providing React components in a config file

​
Backend features

  • Store session data in a simple, flexible array-based format
  • Validate data via schemas
  • Manage session lifecycle via runs
  • Track session versions and prevent incompatible changes after agent updates
  • Expose a clean, stateful API with streaming
  • Manage agent users and their sessions
  • Expose a public, read-only API that can be called directly from the browser
  • Communicate via TypeScript and Python SDKs
All AgentView configuration lives in code.