Written for you
Describe the role in a sentence. The AI proposes a name, description and welcome message, writes the system prompt, suggests starter prompts and draws an avatar.
AI agents that do the work.
Describe the role and UP-TO-DATE Agents writes the rest. Give each agent a model and tools, test it without side effects, follow every run step by step, and put it inside your own apps.
13
Built-in tool groups
4
Ways to embed agents
17
Agent languages
MCP
In and out
Most AI tools give you one chat box. Agents gives you a team of specialists. Each one has its own role, model and tools, runs in a safe test mode until you let it act, and leaves a record of every step it took.
Capabilities
From a one-line brief to a working agent.
Build, test and run agents in one place, then use them wherever your team works.
Your choice of model
Run each agent on Claude, GPT or Gemini, directly or through OpenRouter. Set a writing style, a step limit and which inputs it accepts.
Built-in tools
Grant tools per agent: calculator, Wikipedia and arXiv, charts, web browsing, web and Google search, YouTube, images, video, code, documents and spreadsheets.
Connect MCP servers
Add your own MCP servers in Settings and pick which of their tools each agent can use. UP-TO-DATE Automation can be connected the same way.
Safe test mode
Test chats simulate actions by default, so you can see what an agent would do without it doing it. Real actions only run on agents you allow.
Every run on record
Each run keeps its status, steps, tool calls, tokens and output. Filter the history, cancel a run in flight or retry a failed one.
Delegation you approve
Let an agent hand parts of a task to other agents. It proposes a plan first and only runs it after you reply.
Share and reuse
Keep an agent private or open it to everyone signed in, share it with a person to use or edit, publish it as a template, and import or export agents as JSON.
Embed in your apps
Put the agents screen inside another app as a web component, an iframe or server-driven UI, or as a React library inside UP-TO-DATE apps. Docs and a playground are built in.
Platform
Made to plug into everything else.
The same agents are reachable from the app, other AI clients and your own code.
An MCP server of your agents
Your agents are exposed as tools on an MCP endpoint, next to tools to list, read, create, update, duplicate and delete agents.
API keys and OpenAPI
Create API keys in Settings to call agents from your own servers. The REST API is described by an OpenAPI document.
Seventeen languages
Give an agent several languages and a default. Its name, description and welcome message are translated for each one.
File outputs
Documents, spreadsheets and images an agent creates are saved on the agent, ready to preview and download.
Usage and budgets
See estimated AI spend and token usage over time, broken down by provider, model and agent, and set a monthly budget with an alert threshold.
One UP-TO-DATE sign-in
Sign in with your UP-TO-DATE account. No separate password, and every agent and run belongs to a real user.
How it works
Brief it, test it, put it to work.
1
Describe the role
Write what the agent is for. The AI drafts its prompt, description, starter prompts and avatar, and you can edit any of it.
2
Give it tools and test
Choose a model and grant tools or MCP connections. Chat with it in test mode, where actions are simulated, and read each step.
3
Put it to work
Enable it, share it with your team, call it over MCP or the API, or embed it in another app. Every run is logged.
Built for
One agent per job to be done.
Research
An agent that searches the web, reads the pages it finds and checks Wikipedia and arXiv before it answers.
Reporting
Turn data into charts, spreadsheets and Word or PDF documents your team can download.
Content
A writer with your tone set in its writing style, working in each language your audience reads.
Operations
Connect UP-TO-DATE Automation or your own MCP servers and let a trusted agent take real actions.
Product teams
Embed the agents screen in your own product with the delivery model that fits your stack.
AI power users
Use your agents from any MCP client as tools, and manage them from there too.
Stop repeating the same prompts. Build the agent once.
Describe the role, grant the tools, test it safely and put it to work.