# Runtype

> Runtype is an AI-native product platform that lets developers build, deploy, and manage AI agents and flows across multiple surfaces like web chat, Slack, API, email, and MCP.

Runtype is a managed runtime platform built by Travrse Labs, Inc. for shipping production-ready AI features into real products. It bridges the gap between an AI demo and a production-grade experience by providing agents, flows, tool orchestration, model routing, and multi-surface deployment in a single platform. The platform is actively developed and ships as both a hosted SaaS and an open-source chat widget SDK (Persona.js, MIT licensed).

## What It Is

Runtype is an AI product platform — not a framework you configure yourself, but a managed runtime that handles infrastructure so teams can focus on the experience. Builders define agents that reason and act, wire them into multi-step flows with branching and looping, connect them to tools and APIs, and then deploy to any number of "Surfaces" (channels) without rebuilding for each one. Supported surfaces include web chat widgets, Slack bots, API endpoints, email auto-responders, MCP servers, scheduled cron jobs, SMS, webhooks, WhatsApp (coming soon), Discord (coming soon), and Telegram (coming soon).

## How the Surface Model Works

The core deployment unit in Runtype is a **Surface** — one deployed channel that handles unlimited users and conversations. Plans are priced by how many surfaces you deploy to, not by user count. Key surface types include:

- **Web Chat**: Embeddable chat widgets and copilots for any website or app
- **API**: HTTP endpoints for backend automation and data pipelines
- **MCP**: Extend Cursor, Claude, and ChatGPT with custom tools and data
- **Slack**: AI bots for support, standups, and team workflows
- **Schedule**: Cron-based execution for periodic reports and data syncs
- **Agent-to-Agent (A2A)**: Coordinated multi-agent handoffs using a JSON task protocol

Each surface can have its own auth, routing, and configuration. Adding a new channel means adding a surface, not rebuilding the agent.

## Platform Architecture

Runtype describes itself as "a runtime, not a framework." The platform runs globally on the edge with zero cold starts and includes:

- **Agents**: Autonomous reasoning and planning units
- **Flows**: Visual orchestration for multi-step workflows with branching and looping
- **Tools**: Connectors to any API, MCP server, or custom function
- **Product Evals**: End-to-end experience quality measurement (not just model accuracy)
- **Models**: Any provider — OpenAI, Anthropic, Google Gemini, Meta, xAI, Mistral, DeepSeek, Perplexity, Groq, Cohere, and open-source models — with free switching
- **Protected Parameters**: Credentials are hidden from the model and injected at execution time

The platform also ships an MCP server (`https://api.runtype.com/v1/mcp/protocol`) that can be installed in Claude Code, Cursor, Codex, VS Code, Windsurf, v0, Bolt, Lovable, Replit, and other AI-native tools.

## Persona.js: The Open-Source Chat Widget

Alongside the commercial platform, Runtype publishes **Persona** (`@runtypelabs/persona`) — a themeable, pluggable AI chat widget for websites built in TypeScript with zero framework dependencies. It is MIT-licensed and available on GitHub under `runtypelabs/persona`. As of June 2026, the latest release is v4.4.2. Key capabilities include:

- SSE-based streaming chat with pluggable parsers
- Voice input (Web Speech API or Runtype WebSocket service) and text-to-speech output
- Multi-modal content: text, images, PDFs, documents
- Tool call visualization with human-in-the-loop approval gates
- Artifact side-panel for markdown and component rendering
- 14 render hook plugin system for full UI customization
- Three primary layouts: floating launcher, docked side panel, fullscreen
- Works with any SSE-capable backend (Vercel AI SDK, OpenAI Agents, LangGraph.js, Hono, Express, SvelteKit, and more)

## Setup Path

Runtype offers four entry points: MCP server install, CLI, SDK, or direct API. The MCP path is the fastest — one terminal command adds the hosted MCP server to any compatible AI tool, and OAuth handles authentication without copying API keys. From there, an AI agent can scaffold an agent, wire flows, configure surfaces, and return embed code.

## Update: Persona v4.4.2

The latest published release of the Persona widget is v4.4.2, published June 24, 2026. The repository was last pushed on June 26, 2026, indicating active development. The GitHub repository has 53 stars and covers topics including agentic-ai, webmcp, and vanillajs. The broader Runtype platform changelog is available at runtype.com/changelog.

## Features
- Multi-surface deployment (web chat, Slack, API, email, MCP, SMS, webhooks)
- Visual flow orchestration with branching and looping
- Autonomous AI agents with reasoning and planning
- 100+ model integrations (OpenAI, Anthropic, Google, xAI, Mistral, DeepSeek, etc.)
- Protected parameters that hide credentials from the model
- Product Evals for end-to-end experience quality measurement
- MCP server for use with Claude, Cursor, Codex, and other AI tools
- Agent-to-Agent (A2A) protocol for coordinated multi-agent handoffs
- Scheduled cron-based automation
- Bring Your Own Keys (BYOK) on higher plans
- Persona.js open-source chat widget (MIT)
- Streaming chat with SSE
- Voice input and text-to-speech output
- Tool call visualization with approval gates
- Zero cold starts on global edge runtime

## Integrations
OpenAI, Anthropic, Google Gemini, Meta, xAI, Mistral, DeepSeek, Perplexity, Groq, Cohere, Slack, GitHub, Linear, Google Workspace, Stripe, Vercel AI SDK, LangGraph.js, OpenAI Agents SDK, Hono, Express, SvelteKit, Next.js, Cloudflare Workers

## Platforms
WEB, API, CLI, DEVELOPER_SDK

## Pricing
Freemium — Free tier available with paid upgrades

## Version
@runtypelabs/persona@4.4.2

## Links
- Website: https://www.runtype.com
- Documentation: https://docs.runtype.com/
- Repository: https://github.com/runtypelabs/persona
- EveryDev.ai: https://www.everydev.ai/tools/runtype