We've been writing about the thesis behind this for a while now. The short version: the life sciences industry is at an inflection point. AI capabilities double every 196 days. Legacy vendors are responding by building walled-off models trained on your data — a process that takes 6–9 months and locks you in before you've even started. That's the wrong answer.

The right response is a membrane, not a wall. A protective layer with selective permeability that lets data interact with an evolving ecosystem — without allowing non-compliant actions. That's what we built.

What Kivo Headless is, specifically

Kivo Headless is three things shipped together:

An MCP service — compatible with Claude, Cursor, and any agent framework that speaks the Model Context Protocol. Point your agent at the Kivo MCP endpoint, authenticate with your Kivo credentials, and your agent can query document inventory, read metadata, check workflow state, and understand the structure of your regulated content. It's read-only at launch, which is intentional — more on that below.

A structured action handoff pattern — when an agent has analyzed something and wants to hand a proposed action to a human, it generates a deep link into Kivo that opens the UI to exactly the right context: the document to review, the task to complete, the discrepancy to resolve. The human sees what the agent found, makes a judgment call, and takes the action under their own authenticated identity. Kivo records the action, the actor, and the timestamp.

A skills suite — out-of-the-box skills that give AI tools the necessary context for your specific programs, regulatory requirements, and agent response criteria. Skills can be managed via a change control process and selectively enabled across your organization. You can use them as-is or build on top of them.

AI does the analysis. The human does the work. Kivo records both. That's the architecture.

Why read-only first?

The answer is both regulatory and philosophical.

Regulatory: 21 CFR Part 11 and ICH E6(R3) require that electronic records be attributable to a specific individual. An agent writing a record on behalf of a user isn't the same as a user taking a deliberate action. We don't think the FDA will update their thinking on this quickly, and we're not going to ask our customers to be the test case.

Philosophical: we genuinely believe the human action is part of the value, not a limitation we're working around. The agent makes the human faster and smarter. The human brings judgment, accountability, and signature authority. Taking the human out of that loop doesn't make the workflow better — it makes the audit trail weaker and the outcome less defensible.

One last thing

If you build something interesting with Kivo Headless, we want to hear about it. Reach out at headless@kivo.io

Thanks for being part of this launch. See you at the webinar.