Builder Studio Runtime

Live audit runtime with streaming actions, terminal output, and file-context aware chat.

  • builder-studio
  • runtime
  • sse
  • terminal
  • context-chips

Builder Studio is Sentinelayer's live audit workspace for repo-aware runs.

Runtime UX

  • run timer from `run_started` to completion
  • action badges (`READ`, `SEARCH`, `EDIT`, `TEST`, `SCAN`, `SUBAGENT`, `RECONCILE`, `PR`)
  • collapsible timeline with nested activity rows
  • terminal stream for command output and failure diagnostics
  • auto-collapsed completed traces with final run summary pinned

Chat and context behavior

  • typing indicator replaces generic processing labels
  • selected files are represented as file-context chips in chat
  • context chips track loaded status so you can confirm what the model can reference
  • footer reminder: "AI can make mistakes. Verify important changes."

Runtime safety modes

  • `advice` and `audit_readonly` for non-mutating analysis
  • `edit_gated` and `autonomous_gated` for guarded changes and PR flow
  • approval checkpoints are required before push/PR actions in gated modes

Structured Answers

How do I know what the AI is doing during a run?

Use the live timeline badges and terminal stream. Every meaningful action is emitted as an event and shown in run order.

Can Builder Studio see the file I selected?

Yes, when the file is attached as a context chip and marked loaded. The chip state is the source of truth for model-visible file context.