Thesys

Operator's take

Thesys sits at a specific seam in AI product development: the gap between a model producing good output and that output actually rendering as a usable UI. Teams typically paper over this with a hardcoded layout — a fixed set of cards or tables — and the interface gets increasingly awkward as the model's outputs vary. Thesys bets that the model should own the UI layer directly: the OpenUI Cloud API (formerly called the C1 Generative UI API) lets an LLM specify which components to render, with what data, in what arrangement, and the frontend responds dynamically. It's aimed at developers building AI copilots, dashboards, or chat surfaces where the interface needs to adapt to model output rather than constrain it.

Thesys now ships three products. OpenUI Cloud (formerly the C1 API) is the developer API layer for wiring generative UI into your own stack — the model streams live components (dashboards, forms, charts, slides, reports) as it responds. Reports API is a sibling developer API aimed at durable artifacts rather than streaming UI: it turns conversations, files, and business data into editable slides and reports, template-based and brand-customizable, exported as PDF or PPT. Agent Builder is the no-code platform on top — build and deploy interactive AI agents without writing code. OpenUI Cloud is the flagship and what the generative-UI pitch is really about; Reports API covers the document-generation case; Agent Builder is the on-ramp for teams that want the output UI without owning the integration themselves.

The practical fit for the API is narrow but clear: if you're already building with React and a hosted LLM, and the "static layout doesn't match dynamic output" problem is the blocker, Thesys addresses it directly. Teams that haven't hit that wall yet — or who are building non-AI products — will find it premature.

Pricing is now public and self-serve across all three products. OpenUI Cloud has a real Free tier (3K API calls/month), then Build at $49/mo (25K calls), Grow at $499/mo (500K calls plus SSO/SAML), and a custom Scale tier that unlocks self-hosting or VPC deployment and bring-your-own-LLM. API calls are bundled in the plan; LLM inference is passed through at provider rates with no markup. Reports API is metered separately — 100 free pages a month, then $0.01/page — and Agent Builder runs its own Free/Build/Grow ladder.

What it's good at

• LLM-driven component rendering — the model specifies which UI components appear and with what data; the frontend updates without additional frontend code for each case.
• Real-time state sync — streams UI components as model output flows in; low-latency rendering with stateful endpoints that preserve context and UI state across interactions.
• MCP and tool integration — turns outputs from APIs, databases, MCP servers, and tools into live UI; the API layer handles data integration without extra frontend wiring.
• React SDK with design-system hooks — bring your own React components and design system; customizable components slot into existing patterns without breaking visual consistency.
• Reliability layer — built-in validation, error correction, safe rendering, and model fallbacks so the UI stays stable even when the model fails, slows, or returns partial output.
• Agent Builder (no-code) — a separate no-code product for building and deploying interactive AI agents without API work; same generative UI output, no code required.
• Document and slide generation (Reports API) — a third product: a developer API that turns conversations, files, and business data into editable reports and slides, template-based and brand-customizable, exported as PDF or PPT.

What it's not

• Not for non-AI products — the entire premise is model-driven UI; static sites, traditional web apps, and anything not built around LLM output are outside scope.
• Not a pure no-code tool at the API level — OpenUI Cloud requires React experience and API familiarity; the value prop reduces frontend coding, not the need for developers entirely. Agent Builder is the no-code path.
• Not a category long-shot anymore — the platform now carries SOC 2, ISO 27001, and GDPR certifications and cites 10,000+ users; the "early-stage bet" framing no longer holds.