Taskriz and AG-UI
Both connect AI agents with people. AG-UI is an open protocol that streams an agent’s work into the app a person is already using. Taskriz reaches the person who isn’t in your app, wherever they are, and brings their answer back.
Different jobs
AG-UI
An open, event-based protocol, started by CopilotKit, for the connection between an agent and a user-facing app. The agent’s messages, tool calls and shared state stream to the app live, and the app can pause the agent for the person to approve or edit. You build the app; AG-UI is how it talks to the agent.
Taskriz
A service for getting things from people. Your software, your Agent or you send a request; Taskriz takes it to the right person, by email, text message, a link, Slack, Google Chat or a request box on your website, and brings their answer back as structured data. The people who answer need no app and no account.
Use them together
They fit together. An agent that streams to your app through AG-UI can call the Taskriz MCP server or the Taskriz HTTPS API when it needs someone outside your app: a customer’s approval, a contractor’s photo, a supplier’s price. It can pause its run with an AG-UI interrupt while it waits, show the person in your app that it’s waiting, and carry on when the answer comes back, or have Taskriz call your webhook.
Your team can also ask from where it already works: Slack, Google Chat and Microsoft 365 Copilot. And visitors to your website can send you a complete request with Website Requests.
What they share
- A human in the loop. AG-UI lets your app pause an agent for a person; Taskriz brings a person into an agent’s work when it needs them, with reminders and text messages if they don’t answer.
- Interfaces made for the moment. AG-UI carries typed components for your app to show; Taskriz sends a whole screen built for the answer, run in a sandbox on the person’s device.
- Any agent. AG-UI has adapters for many agent frameworks; Taskriz works with any Agent or software through plugins, MCP or its API, and with none at all through Ask Someone.
Side by side
| AG-UI | Taskriz | |
|---|---|---|
| What it is | An open, event-based protocol (MIT licence) for connecting an AI agent to a user-facing app, started by CopilotKit with LangChain and CrewAI | A hosted service for reaching people, with an app, plugins, an MCP server and an HTTPS API |
| Which way it works | The agent’s work streams into the app the person is using: messages, tool calls, state and progress, live | An agent, your software or you ask a person for something, and get their answer when it comes |
| Who answers | The person using your app, while the agent runs | Anyone a request is sent to, from the link they’re sent: no app and no account, straight away or days later |
| Reaching people | Not part of it: the person is already in your app | By email and text message (Australia and the United States), a link, Slack or Google Chat, or a request box on your own website |
| Telling them a request is waiting | Not part of it | In the Taskriz app, by email, browser notifications and text message, with reminders if no one answers |
| Where people answer | In your app’s own interface, which you build and run (for example with CopilotKit, or in Slack and Teams through CopilotKit’s clients) | The Taskriz app in the browser on any phone or computer, nothing to install; in Slack and Google Chat; or on your website |
| What a request can ask | Whatever your app shows: text, tool results, typed components, and an interrupt to pause for the person to approve, edit or retry | A whole screen built for the answer: choices, dates, a photo or document, a map or a short note, from iOS-style building blocks |
| Answers back to the Agent | In the same live connection; after an interrupt, the app resumes the agent with a new run | As structured data, matched to the request: through the plugins, the MCP server, a signed webhook, a WebSocket or the events list |
| Connecting your Agent | Adapters for agent frameworks (LangGraph, CrewAI, Microsoft Agent Framework, Google ADK, Mastra, Pydantic AI, AWS Strands), Amazon Bedrock AgentCore, and SDKs in many languages | Plugins for Claude and ChatGPT, the Taskriz MCP server for any MCP client (Copilot Studio included), and an HTTPS API for any software, Agent or workflow. No AI of your own? Ask Someone writes the request from your plain words |
| People starting things | The person types to the agent in your app | Your Agent’s request catalog: actions people start from the Taskriz app, and visitors to your website through Website Requests |
| Cost | Free and open source; you build and run your app and your agent | A Free plan, then paid plans by the requests your Agents send; the people who answer never pay |
Security
AG-UI is a protocol, so its protections are your app’s: its sign-in, its connection and how you run your agent. In Taskriz, a request’s protections come built in.
| AG-UI | Taskriz | |
|---|---|---|
| Who can reach a person | No one is reached; your app’s own sign-in decides who sees the agent | Only people your Agent sends a request to, or who paired with it. Someone without an account opens a request with one tap on their first device (a one-time code elsewhere), and anyone can decline, or stop an Agent reaching them |
| Encryption | Your app’s connection, over HTTP or WebSockets | Each request is sealed to the person’s own devices and stored that way, and travels over TLS 1.2 or later |
| Who can read a request | Your app, your agent and the AI model it calls | Only the person’s devices: it passes through Taskriz while it’s delivered, and is never stored or logged readable |
| Running the interface | Your app’s own code, in its own pages | In a WebAssembly sandbox with no network or file access of its own |
| Confirming what matters | Your app can pause the agent with an interrupt, for the person to approve or edit before it goes on | The person sees the whole request and answers it themselves |
To be plain about Taskriz too: files sent with a request or reply are encrypted on our servers but not sealed to devices, and two-step sign-in is on the Roadmap. Security has the details.
When Taskriz fits
- The work is waiting on someone who isn’t in your app, such as a customer, contractor or supplier, and they shouldn’t have to install or sign up for anything.
- The answer may take minutes or days, and your agent or software should pick it up when it comes.
- Your team asks from Slack, Google Chat or Microsoft 365 Copilot, or you have no AI of your own and want to ask in plain words.
- You’d rather not build and run an app, notifications, sign-in and delivery to get an answer back.
When AG-UI fits
- You’re building an app with an agent inside it, and want its work to stream to the person live.
- The person is in your app while the agent runs, and approves or steers it there.
- You want an open protocol with adapters for the agent frameworks your engineers already use.
Questions
What’s the difference between AG-UI and Taskriz?
AG-UI is an open protocol for connecting an AI agent to the app a person is using: the agent’s messages, tool calls and state stream into that app, live. Taskriz is a service for reaching a person who isn’t in your app: your Agent sends a request, Taskriz takes it to them by email, text message, a link, Slack or Google Chat, and brings their answer back as structured data, minutes or days later.
Is AG-UI the same as A2UI or MCP?
No. AG-UI connects an agent to a user-facing app; MCP connects an agent to tools and data; A2UI is a specification for the interface an agent describes. Taskriz works through MCP: the Taskriz MCP server gives an agent tools to send people requests and read their answers.
Can AG-UI reach someone who isn’t using my app?
No, it isn’t made to. AG-UI carries an agent’s work to the app the person already has open; it doesn’t find, notify or reach anyone else. Reaching a customer, contractor or supplier who isn’t in your app, and bringing their answer back, is what Taskriz does.
Can an agent that uses AG-UI also use Taskriz?
Yes. An agent that streams to your app through AG-UI can call the Taskriz MCP server or the Taskriz HTTPS API like any other tool, when it needs someone who isn’t there. It can pause its run with an AG-UI interrupt while it waits, and carry on when the answer comes back.
Who supports AG-UI?
Amazon Bedrock AgentCore Runtime announced AG-UI support in March 2026. AG-UI’s own documentation lists first-party support from frameworks including Microsoft Agent Framework, Google ADK, LangGraph, CrewAI, Mastra, Pydantic AI and AWS Strands, and community-built adapters for the Claude Agent SDK and the OpenAI Agents SDK.
About this comparison
AG-UI was checked in November 2026, from the AG-UI documentation, its events, AG-UI on GitHub and Amazon’s AgentCore announcement. AG-UI is an open-source project started by CopilotKit; Taskriz is not affiliated with it. What Taskriz is building next is on the Roadmap.