intervu is a local, AXI-style CLI that turns an agent-generated HTML artifact into a collaborative browser review surface: the human annotates the rendered page, and the agent long-polls for that feedback and live-edits the file.
See CONTEXT.md for the ubiquitous language and
docs/adr/ for the architectural decisions.
The review loop runs by having your agent shell out to a bare intervu command, so intervu has to be on your PATH.
intervu ships as a single self-contained Bun binary, so you need Bun on your PATH too.
Install it globally and wire it into your agent:
bun add -g intervu
intervu setupintervu setup installs the agent skill and the session-start hook that make the review loop discoverable.
It refuses when intervu is not yet on your PATH, because the skill and hook it writes both invoke a bare intervu - a transient bunx run would leave them pointing at a command that does not exist.
For a one-off manual look at an artifact you can run it without installing, but this is not enough for the agent loop:
bunx intervu report.htmlInstall dependencies:
bun installRun the dev server (watches and restarts on file changes):
bun run devOpen an artifact for review - launches the browser chrome around it:
intervu open report.html
# bare-file alias:
intervu report.htmlLong-poll for human feedback; blocks until the human acts, then returns the queued feedback as TOON. Safe to kill and re-run with no loss:
intervu poll report.html
# post a reply into the human's conversation panel as you re-poll:
intervu poll report.html --agent-reply "tightened both headings"bun run format # biome format --write
bun run lint # biome lint --write
bun run check # biome check --write (combined + import sort)
bun run check:types # tsc --noEmit (effect-tsgo-patched)
bun run test # vitest runReleases are automated with Changesets. To record a change for the next release:
bun run changeset # describe the change and pick a semver bump