Paimen (pronounced “PIE-men”) is Finnish for shepherd — the one who keeps the flock together, watches over it, and makes sure nothing strays. That’s what Paimen AI does for your AI agents: components stay on the allow-list, agents act within your rules, and every decision leaves a traceable record.

Yes — Rebel AI Studio is now Paimen AI. Same platform, same team, same roadmap; the name simply fits a product built for control better. Read the full story →

Agentic UX is what you get when agents — not fixed screens — assemble what each user needs, moment to moment, within rules you set. “Agentic UI” is the mechanism (agents generating interface); agentic UX is the outcome — a product that adapts to intent instead of making people navigate to it. Paimen AI is the platform for building and governing it: generative UI under the hood, allow-lists and an audit trail around it.

Three industry terms describe what Paimen AI is built on:

GenUI (Generative UI) — an LLM generates UI elements at runtime instead of picking from a static layout. Paimen uses it to assemble the right interface for each intent, from your existing design system.

Agentic UI — the design paradigm where autonomous agents proactively drive the interface, mixing text replies with real, interactive components.

A2UI (agent-to-UI) — Google’s open declarative protocol for streaming UI from an agent to a native client, rendering safely without executing arbitrary code. Paimen is A2UI-aligned.

Paimen implements all three and adds the layer that makes them shippable: allow-lists, versioning, audit trails, and analytics.

Business owners edit prompts, FAQ entries, and conversational guardrails through the management console — no code required. Engineering owns technical components and integrations. Role-based permissions enforce the split, and every change is attributable to a real person in your identity provider.

Edit and test in an environment separate from production, run new prompt versions side by side against the same questions (blind comparisons supported), then deploy to a limited group before rolling out widely. Every change is versioned and reversible at any time.

Yes. Every recorded interaction is tagged with the exact prompt version that was active at that moment, so post-hoc reviews show what the AI did, what the user saw, and which rules shaped the outcome.

No. It uses your existing design system and components. Think of it as a layer that decides which components to show, when, and why.

Yes. Guardrails, policies, and audit trails ensure safe behavior. Every agent action and component is allow-listed by your team.

LLMs from leading providers: Google Gemini, OpenAI models, and more. The platform is model-agnostic.

Built-in analytics aggregate real end-user interactions across the deployed application. The console surfaces what users were trying to do, which components were used, and where their path broke. Reports identify recurring themes and suggest next steps, and each finding links back to the original sessions — verifiable in the source data, not an AI summary. Native BigQuery export keeps unlimited history for deeper analysis.

Yes — that's a core use of the platform. Paimen brings AI agents into the same management console, where your team can monitor them. Every agent decision is auditable, tied to a version, and recoverable when needed.

No. AI interprets intent, but the components themselves remain traditional, well-tested code from your design system.

Yes — the platform's primitives map directly to several of the obligations the EU AI Act places on high-risk AI systems.

Transparency (Articles 13 and 50). Components rendered to the user come from your allow-listed catalog. The agent can only request approved components, and the application renders them from your design system, so what users see is always inspectable code. Answers that draw on data cite the original records they came from. AI-mediated interactions identify themselves to the user; they don't pretend to be plain UI. Every output is traceable back to the exact prompt version that produced it.

Record-keeping (Article 12). Per-interaction audit trails tagged with the active prompt version are the platform default.

Human oversight (Article 14). Allow-lists, role-based controls, and agent decisions reviewable from the console.

Quality management (Articles 16–17). Versioning, rollback, and side-by-side prompt testing.

Paimen doesn't make you compliant on its own — your risk classification, conformity assessment, and post-market monitoring remain your obligations — but it gives you the technical foundation the Act effectively expects.

Three structural answers. First, the self-hosted deployment option keeps personal data inside your own infrastructure and authentication boundary — the platform never has to leave your stack. Second, allow-lists and permissions limit which data and components are made available to the agent, so data minimization happens at the architecture layer rather than as a policy hope. Third, the per-interaction audit trail (with IdP-bound user identity) is the evidence layer for access-control and right-to-erasure requests. As with any platform, your organization remains the controller and is responsible for lawful basis, retention, and DPIA decisions.