B4.run product-loop media transcript

The flagship and three derivative clips are silent. Their visible short act labels — Author, Prove, and Run — are repeated here so the same proof remains available without motion or audio. The footage begins inside an existing generated research workspace; it does not show the scaffold command running.

Product loop (25 seconds)

Author

The B4.run capture compositor displays real files from the generated research workspace. Its tree names all five paths used by the story:

• server/src/app/research/index.ts
• server/src/app/research/state.ts
• server/src/app/research/plan.md
• server/src/tools/searchCorpus.ts
• server/test/research.test.ts

The editor focuses on the export default agent({ route descriptor and the shared searchCorpus tool. The compositor uses frozen-frame holds to keep the real captured source legible; the holds do not add product events.

Prove

The terminal compositor runs the generated workspace's canonical npm test command. The real, narrowly normalized log shows the named deterministic research scenario passing without a provider key. A frozen-frame hold leaves the passing result on screen long enough to read.

Run

The actual generated B4.run Workbench submits “What are common agent architectures?” to /research#agent. Visible tool activity names the searchCorpus and readDoc calls. The cited response reads: “ReAct and plan-and-execute are common. [corpus/agent-architectures.md]”.

The completed run remains visible, then the browser reloads. The same thread is reopened from the rail, its state request succeeds, and the prompt, both tool calls, and cited answer reappear from the server checkpoint. This demonstrates browser-reload restoration with the B4.run server still running, not restoration after a server restart.

Close

The closing card reads “TypeScript meta-framework for LangGraph.js”, “Build LangGraph agents like Next.js apps”, and npm create b4-app@latest my-agent. The command is an activation next step; scaffolding is not part of the footage.

Author clip (9 seconds)

The Author derivative holds the same real generated file tree, route descriptor, and shared tool on screen for a legible static inspection.

Test clip (9 seconds)

The Test derivative shows the same real npm test result and named offline research scenario, with a legibility hold after the passing summary.

Run clip (10 seconds)

The Run derivative begins with the completed Workbench response, follows the browser reload and same-thread selection, and ends with the restored prompt, searchCorpus and readDoc activity, and cited answer.