FlowOS overview, the AI Operating System for your company.

FlowOS is not just an assistant. It is an operating system for AI-assisted work: business agents, agent chains and scheduled tasks, measurable and auditable, that prepare the work while your teams keep the decision.

Since 2023, after months in a research lab building internal prototypes for pilot partners, we are opening the platform today. Here is how it works, why we made some decisions that may surprise you, and how a deployment works.

The problem we solve

Companies generate a growing volume of repetitive, contextual, measurable work: meeting minutes, customer follow-ups, meeting prep, document analysis, reporting. This work is too voluminous to stay entirely human, too nuanced for fixed rules, and too sensitive for a simple prompt to an LLM.

Existing solutions fall into one of three categories:

• Chatbots that answer politely but trigger nothing
• No-code automations that break down at the first edge case
• Copilots that help a human do the work, without ever doing it for them

None of them prepares business work end to end, reviewed and approved by your teams. FlowOS is built for that.

What FlowOS does

FlowOS orchestrates specialized AI agents, each with a clear scope, consumption caps and human review. You describe the task in natural language, and FlowOS prepares the deliverable.

For example, for meeting minutes:

# flowos.config.yaml (example)
agent: meeting-minutes
source: teams-transcript
deliverable: "minutes, decisions and actions"
then: draft-email-to-attendees
cap: per agent
review: before anything is sent

From there, the agent reads the transcript, writes the minutes with decisions and actions, then prepares the draft email to attendees. Nothing goes out without review, and the request is recorded in the audit log.

Four decisions that may surprise you

1. No generalist agent.

We refuse to sell an "agent that does everything." Each FlowOS agent has a narrow scope and a single success metric. This constraint allows three things: objectively evaluating whether the agent performs, quantifying ROI precisely, and limiting the damage when something goes wrong.

2. Escalations are not a failure.

At some competitors, an agent that escalates is considered an agent that "failed." For us, it is expected behavior. The goal is not to reach 100% autonomy, it is to reach an autonomy rate that is predictable, measurable, and compatible with your risk tolerance.

3. Budgets are hard limits.

Each agent can have a spending cap, checked before every model call. When it is reached, the agent stops, before the bill arrives. It is the only way we know to make an AI deployment financially predictable.

4. Every action leaves a trace.

Important actions (sign-ins, documents, shares, model calls, agent usage) are recorded in an audit log. In practice: you know what your AI did, when, and at whose request.

What FlowOS does today

The use cases FlowOS covers today:

• Meetings, minutes from Teams and Zoom transcripts, with a draft email to attendees
• Email, reading emails and drafting replies in Outlook and Gmail, never sent automatically
• Documents, proposals, memos, contract and tender analyses, in Word, PowerPoint and Excel
• Slack and Teams, agents called in channels, scheduled tasks posted to a channel
• FlowOS ERP, customer context within permissions, reminders, meeting prep and the morning briefing

In every case, the deployment follows the same method: a narrow scope, goals set together, a pilot, then a deployment decision.

What FlowOS does not do

We also need to be clear about what it is not FlowOS:

• No creative marketing content generation
• No open-ended scientific reasoning
• No replacing roles that require human judgment on ethical, strategic, or sensitive HR matters
• No deployment without a human-in-the-loop period

If you are looking for a "smart chatbot," FlowOS is oversized. If you are looking for a general-purpose AGI, it does not exist yet, and we would rather tell you so honestly.

What's next?

FlowOS is available in three plans: Starter, Team and Enterprise (with private deployment possible), all custom-quoted. All plans include the audit log and support in French or English; hosting is defined with each customer.

To go further:

• Read the full product page
• View detailed pricing
• Request a 30-minute demo

We don't sell a vision: we show FlowOS on your own documents. For usage examples, our illustrative scenarios are available here.

— The Thinking Platforms team