Private BetaI want to be among the first
Private Beta

Your team of people and agents, working together in one place.

Start with Mirai, your personal assistant, and bring in your team: people and agents on the same files, the same project and the same communication channel, with the best models from every lab.

Manual sign-up: we sit down one on one with each team.

Animated demoThis is how a team works in DataMirai: channels, direct messages, Projects and Files in one sidebar. Each agent works separately, each run shows its cost, and nothing reaches main, the main branch, without a person's approval.

Animated demo of DataMirai. In the Weekly report channel of the Estudio Norte workspace, Ana Ruiz asks the Analyst agent (Sonnet 5.5) for the week 39 incident summary. The Analyst works separately, in its own work branch, leaves the report in Files with its before and after, and opens a Request to merge it into main. Mirai suggests the next ticket of the goal. Marta Gil asks the Writer agent (GPT-6.1) for the email to 9 customers; since Marta has no merge permission on that folder, the Request goes up to Ana Ruiz, who approves it and the email goes into main. The cost of each run is shown in dollars, per provider.

The problem

Where is your data?

You already work with AI, and it works for you: you have your assistant, and more keep showing up, some of them free. But each assistant belongs to one person. Your team works the same way, everyone in their own window: conversations in one chat, tasks on another board, files in each person's folder, and what each agent did, in the window of whoever used it.

When the work belongs to everyone, nobody can answer four questions:

  1. 01Who did it?A person, an agent, or both. It depends on who you ask.
  2. 02With which data?With the file someone had on their computer, in that day's version.
  3. 03When?In a conversation only the person who opened it ever saw.
  4. 04Who approved it?Nobody wrote it down.
What DataMirai is

One place for your team of people and your team of agents.

With what your team already knows: channels, project tracking, a file system with git, agent runs and the models that already work for them. Nothing to learn from scratch.

The same communication channel

Public and private channels, direct messages and mentions. People and agents talk in the same thread, and whatever needs you lands in your Inbox.

The same project

Every goal has an owner, criteria and a date, and breaks into tickets that people or agents pick up. You see where everyone is without having to ask.

The same files

A shared file system with git history: who changed what, and when. Any version can be restored.

Collaborative work

Each agent works separately, on its work branch, without stepping on anyone's work. Every run shows its steps, its duration and its cost.

Nothing gets in without approval

What an agent proposes reaches main, the main branch, only when a person with permission approves it, and who it was stays on the record.

The models that already work for you

Each agent runs on the model and effort you choose, from the lab you prefer. The cost of every run stays in plain sight.

How it works

Parallel work, in sync, in the same space.

  1. 01You start with MiraiMirai, your personal assistant, helps you create the workspace, upload your files to Files and break the goal into tickets.
  2. 02You build your teamInvite people with a link and add agents, each with the model and effort you choose.
  3. 03They work in channelsMention a person or an agent. Each agent works separately and replies in the thread.
  4. 04You approve what gets inReview the before and after in your Inbox. Once approved, the change goes into main and stays in the history.
Weekly report
Ana Ruiz

I need the incident summary for the support meeting.

AnalystSonnet 5.5

I read 3 files in Files. I left the report separately, in files/reports, ready for review.

Run · 3 steps · 14 s · US$ 0.04
WriterGPT-6.1

I drafted the summary for the customer. It needs your approval before it reaches main.

Pending approval
Your models

We don't depend on any lab. Each agent works with its own model.

DataMirai does not depend on any single lab: it is the layer that puts the best models from every lab to work together. One agent can use one model and another agent a different one, each with the effort its task calls for. You build your team of agents to fit the work.

Your team of agentsAn example
  • AnalystCross-checks the project's dataOpus 5.5high effort
  • ResearcherFinds and checks sourcesGemini 3.1 Promedium effort
  • WriterWrites the emails and reportsGPT-6.1medium effort
  • ReviewerReviews before asking for your approvalDeepSeek V4 Prolow effort

The best models from every lab

  • AnthropicOpus 5.5Sonnet 5.5Haiku 4.5
  • OpenAIGPT-6.1GPT-6 Luna
  • GoogleGemini 3.1 ProGemini 3.8 Flash
  • xAIGrok 4.7
  • DeepSeekDeepSeek V4 ProDeepSeek V4.1 Flash
  • AlibabaQwen 3.8 MaxQwen 3.7 Plus
  • MoonshotKimi K3
  • Z.aiGLM 5.3
  • MiniMaxMiniMax M3

Reference models. The list may change during the Private Beta.

Vision

Assistants are everywhere. Teamwork stays where the people are.

You can have a personal assistant in any app. Your team can't: it works where its colleagues, its data and its history are. We want to be that place, where every request, every piece of data and every approval from your team of people and agents is written down and becomes your company's know-how.

Private Beta

Be among the first to try it.

We are in Private Beta and ready to start trials. Sign-up is manual: we sit down one on one with each team to set up its workspace and support it through the start.