Top AI Repos — open-source AI, indexed and scored
Top AI Repos tracks AI repositories on GitHub and answers two different questions about each one: is it moving right now, and would you bet a product on it.
Top AI Repos tracks AI repositories on GitHub and answers two different questions about each one: is it moving right now, and would you bet a product on it.
Learn how to use CUA (our Computer Using Agent) via the API on multiple computer environments.
| Date | Stars |
|---|---|
| 2026-07-31 | 1758 |
| 2026-08-06 | 1758 |
Today
— stars today
This week
— stars this week
This month
— stars this month
Momentum
0.0
growth rate 0.00%/day
# GPT-5.4 CUA Sample App TypeScript sample app for browser-focused computer-use workflows with GPT-5.4. The repo includes: - `apps/demo-web`: a Next.js operator console for starting runs and reviewing screenshots, events, and replay artifacts - `apps/runner`: a Fastify runner that manages mutable workspaces, browser sessions, SSE, and replay bundles - `packages/*`: shared scenario, runtime, and contract packages that make it easy to add new labs later The legacy Python sample does not ship in this release branch. Keep that history on a separate `v1` or `legacy` branch. ## What This Repo Demonstrates - how to integrate the Responses API from one canonical place: `packages/runner-core/src/responses-loop.ts` - how to switch between `code` mode and `native` computer mode against the same browser lab - how to define scenario manifests, launch isolated run workspaces, and verify outcomes - how to build an operator-facing console that is understandable even when the runner is offline or a run fails ## Prerequisites - Node.js `22.20.0` - pnpm `10.26.0` - Playwright Chromium browser install ## First Run ```bash git clone <repo-url> cd openai-cua-sample-app corepack enable pnpm install cp .env.example .env ``` Edit `.env` and set at least this environment variable: ```bash OPENAI_API_KEY=your_key_here ``` The runner reads the repo-root `.env` automatically when you start it through the provided scripts. The web app uses its built-in defaults; if you need to override `NEXT_PUBLIC_*` settings, add them in `apps/demo-web/.env.local`. If `pnpm install` prints an `Ignored build scripts` warning for optional packages such as `sharp` or `esbuild`, you can ignore it for local development in this repo. A clean clone still installs, builds, and starts successfully without approving those scripts. Install the Playwright browser: ```bash pnpm playwright:install ``` On Linux, install Playwright OS dependencies as well: ```bash pnpm playwright:install:with-deps ``` If Playwright later reports missing system libraries, rerun the `with-deps` command above and follow any OS package prompts it prints. Start both apps together: ```bash pnpm dev ``` Open [http://127.0.0.1:3000](http://127.0.0.1:3000), choose a scenario, keep `Headless` selected, and start a run. ## Local Development Run the services separately if you want independent logs: ```bash pnpm dev:runner RUNNER_BASE_URL=http://127.0.0.1:4001 pnpm dev:web ``` Common checks: ```bash pnpm lint pnpm typecheck pnpm test pnpm build pnpm check ``` Live smoke tests stay opt-in and secret-gated: ```bash OPENAI_API_KEY=your_key_here pnpm test:live ``` ## Execution Modes - `native`: exposes the Responses API computer tool directly. The model requests clicks, drags, typing, waits, and screenshots against the live browser session. - `code`: exposes a persistent Playwright JavaScript REPL through `exec_js`. The model scripts the browser rather than emitting raw computer actions. Both modes use the same scenario manifests and replay pipeline. `native` is the closest sample of the computer tool itself. `code` is the clearest sample of a browser REPL harness. ## Official Scenarios - `kanban-reprioritize-sprint` (`kanban`): teaches stateful drag-and-drop verification against a target board state derived from the operator prompt - `paint-draw-poster` (`paint`): teaches cursor control, drawing, and verifying saved canvas state against the live canvas - `booking-complete-reservation` (`booking`): teaches multi-step browsing and form completion with verification against a local confirmation record More detail lives in [docs/scenarios.md](docs/scenarios.md). ## Repo Map - `apps/demo-web` The operator console UI - `apps/runner` The HTTP runner, SSE endpoints, and artifact serving layer - `packages/replay-schema` Shared request, response, replay, and error contracts - `packages/scenario-kit` Public scenario manifests and prompt defaults - `packages/browser-runtime` Playwright
Excerpt of 5,560 characters
Read on GitHub
Katia Gil Guzman · @openai
15
Ilan Bigio · OpenAI · United States
7
Kazuhiro Sera · OpenAI · Japan
3
Charlie Guo
2
2
Romain Huet · @OpenAI · United States
1
Takahiro Menju · Japan
1
1
1
1
1
Would you bet a product on this? Bounded 0–100 and slow moving.
matched fp:133eef39d4d4a76a, llm:description: 'Learn how to use CUA (our Computer Using Agent) via the API on multiple computer environments.' Language: TypeScript. CUA = Computer Using Agent — agent that uses computer environment via API.
matched fp:133eef39d4d4a76a, llm:description: 'Learn how to use CUA (our Computer Using Agent) via the API on multiple computer environments.' Language: TypeScript. CUA = Computer Using Agent — agent that uses computer environment via API.
matched fp:133eef39d4d4a76a, llm:description: 'Learn how to use CUA (our Computer Using Agent) via the API on multiple computer environments.' Language: TypeScript. CUA = Computer Using Agent — agent that uses computer environment via API.