How
Generation API/SDK: the model outputs UI as OpenUI Lang (or C1 responses) that a client renders; no live control channel.
Hosted generative-UI API plus open compact UI language.
thesys.dev ↗repo ↗docs ↗OpenUI Cloud (formerly C1 API) ↗OpenUI playground ↗
A web generation service; only its compact-language token-cost lesson applies to scene.
Ignore it for this purpose.
Thesys attacks the real cost of generative UI: tokens and latency. OpenUI Lang is a purpose-built compact language that, by its own reproducible tiktoken benchmark, needs roughly half to a third of the tokens of json-render or C1's own old JSON, which means faster first paint and cheaper turns. The framework is MIT, renders in four web frameworks, interoperates with A2UI and AG-UI, and the hosted OpenUI Cloud adds the unglamorous production parts (validation, correction, fallbacks, model gateway, observability) teams otherwise build themselves.
| UI as data | 4 scene 5 | OpenUI Lang is a serializable text format, but not JSON. |
|---|---|---|
| Agent can drive it | 2 scene 5 | More than code generation (constrained DSL rendered safely) but no protocol to change a live instance. |
| Agent can see it | 1 scene 5 | No readback mechanism found in docs reviewed. |
| Small, targeted edits | 1 scene 5 | |
| Time travel | 0 scene 0 | |
| Live data rate | 1 scene 5 | |
| Terminal native | 0 scene 5 | |
| Pixel graphics | 3 scene 4 | Web charts and components. |
| Teaching | 0 scene 3 | |
| Maturity | 3 scene 1 | |
| Openness | 3 scene 4 | MIT framework, proprietary hosted API. |
Thesys C1 (now OpenUI Cloud) and OpenUI scene now scene planned
Generation API/SDK: the model outputs UI as OpenUI Lang (or C1 responses) that a client renders; no live control channel.
Low for web apps: swap the LLM endpoint (C1) or add the OpenUI renderer and prompt.
Not documented beyond normal chat/app state (not verified).
Not documented; responses are regenerated per turn (not verified).
None documented.
Optimised for cheap, fast generation of web UI, not for operating a live interface.
C1 was an OpenAI-compatible LLM API that returns UI (charts, forms, tables, cards) instead of text, rendered by Thesys's React SDK. It is now 'OpenUI Cloud', built on the open OpenUI framework: developers define a component library (Zod-typed), generate a system prompt from it, the model streams OpenUI Lang (a compact text language, not JSON), and React/Vue/Svelte/Angular renderers draw it progressively. @openuidev/a2ui offers an A2UI v1.0 client with OpenUI Lang component payloads; @openuidev/langchain streams OpenUI through AG-UI.
Token benchmark is published in the repo (reproducible); render-speed and user-preference claims are marketing without published method.
Commercial generative-UI API rebranded as OpenUI Cloud, with an open-source OpenUI framework that is growing quickly (10.5k stars).
Used by: Entelligence (Ask Ellie), per thesys.dev, CTRLAgent, FieldCamp (logos on thesys.dev)
Active open-source repo backed by a commercial product.
Releases: Frequent package releases (five packages released 2026-10-08) · Contributors: ~50 (GitHub contributors API incl. anonymous) · Recent: 361 commits on main since 2026-07-11 (GitHub API) · Bus factor: medium: one company
Agent-generated UI constrained to a trusted component library.
Little; scene does not target model-generated UI token cost.
Terminal rendering, live control, readback, data streams bypassing the model, pixels, history.
| 1–7 | home · features · use cases · docs · changelog · under the hood · blog |
| t | light or dark |
| j k | scroll |
| g G | top · bottom |
| ? Esc | this · close |