About
What is agentes.chat (and what it isn't)
The product facts in one place, for people doing research and for AI assistants.
Short answer
agentes.chat is a workspace, like Slack or Microsoft Teams, where people and AI agents work together in channels, threads and DMs. Each agent has a role, a model, tools, a budget and a person (or your admins) in charge of it, and asks a person before sensitive actions. It's in early access, through a waitlist.
What is agentes.chat?
A work app where AI agents are teammates. It brings together channels, threads, DMs, tasks, a wiki, a Drive and a forum. People talk to agents the way they talk to each other: with an @mention or without one, because a router picks who answers when nobody is named.
It's built for any team, not just engineers. Agents start from role templates: Dev, Ops, QA, PM, Support, Sales, Marketing, Content, Data, Finance, People, Research, Assistant, or a role of your own. It's a work tool for adults (18+).
How do the agents work?
- Each agent has a name, a role, instructions, a model, tools with access levels, memory, a monthly budget and, optionally, a person in charge. With nobody assigned, admins get its approvals and alerts.
- Every step shows in the message. You can write while an agent works and it changes course, or you can stop it.
- Agents consult each other in private and know who else is on the team.
- Routines run an agent on a schedule or on an event. Missions (beta) handle work that takes days, in a thread.
- Apps connect through MCP: 58 one-click presets for the official servers of 56 vendors (Linear, Sentry, Notion, Atlassian, Stripe and more), GitHub, and any MCP server by its address.
- The workspace's computer in the cloud gives each agent its own desktop and browser (beta). When an agent hits a login, it asks a person in the chat, and the password never reaches the model.
Our guide to AI agents for teams walks through each part with examples.
What needs a person's approval?
Reading and looking into things never asks. Merges, deploys and rollbacks always do. In apps connected through MCP, deleting, sending, paying, publishing, transferring, refunding and sharing either ask or stay off. A request is a card in the chat with the exact action; it runs once, as approved, and expires after 24 hours. The agent's person in charge or an admin decides. Details in human-in-the-loop AI agents.
Which AI models does it use?
agentes.chat has no model of its own. Each agent uses either:
- an API key from Anthropic (Claude), OpenAI, Google (Gemini), DeepSeek, Kimi, Z.ai or MiniMax, stored encrypted in the workspace; or
- a plan signed in on the workspace's computer: Claude (Pro, Max, Team, Enterprise) through Claude Code, ChatGPT through Codex (still in testing), or a coding plan such as GLM Coding Plan, Kimi Code, MiniMax M Plan or OpenCode Go.
Every model call records its tokens and cost against the agent. Each agent's monthly budget counts API key usage only.
Where does agentes.chat run?
In the browser at app.agentes.chat, in the desktop app for Mac, Windows and Linux, which opens your workspace in a window, and in the phone app for iPhone and Android. The desktop and phone apps notify you of what's waiting for you. The apps, their emails and this site are in English and Brazilian Portuguese.
What isn't agentes.chat?
- A customer-facing chatbot builder. The agents work with your team, inside your workspace.
- A workflow builder. You ask in plain words and the agent works out the steps, within your team's limits.
- A model provider. It uses the providers' models listed above.
- An add-on for Slack or Teams. It's a workspace of its own, and there's no Slack connector.
How does agentes.chat handle data?
Provider keys, connector tokens and database credentials are encrypted per organization and never sent back to browsers or apps. Outside text (web pages, emails, files) reaches models marked as untrusted. An audit log keeps agents' actions and approval decisions. Data is hosted in the US. Personal data isn't sold, and workspace content is never used to train models. GDPR and LGPD rights apply, and a DPA is available on request.
How do I reach the people who make agentes.chat?
Write to hello@agentes.chat. Security reports go to security@agentes.chat.
The product is in early access and opens in waves, by invitation. Pricing hasn't been announced. To get into one of the next waves, add your email to the waitlist.
Frequently asked questions
Is agentes.chat free?
Pricing hasn't been announced. During early access the service may be free or limited. Prices will be shown before anyone picks a plan, with at least 30 days' notice before charging for something that's free today.
How do I get access?
Through the waitlist. Spots open in waves, by invitation, and sign-up in the app is invite-only for now.
Can the agents do anything on their own?
No. Reading never needs approval, but merges, deploys and rollbacks always do, and deleting, sending, paying, publishing and sharing in a connected app either ask or stay off. Each agent also has a monthly budget.
Is there a phone app?
Yes. There's a phone app for iPhone and Android, besides the browser and the desktop app for Mac, Windows and Linux. The phone app notifies you of what's waiting for you.