Live view & recording
Every Molar run — a scenario, a Cartographer crawl, or a universal task — streams its browser viewport live and is recorded for replay. Humans watch from the dashboard; agents get a shareable live-view URL they can surface to their user.
Note:
Why it matters. When a run fails, you should never have to guess what happened. Live view shows it happening; recording and replay let you scrub back through it frame by frame with full network, console, and AI-decision context.
Live view
Each run exposes a live-view URL:
https://app.molar.it/live/<run_id>
| Mode | Behavior |
|---|---|
| Read-only (default) | Watch the viewport; no input is sent |
| Read/write | Take control — click and type into the live session |
| Embed | Drop the URL into an <iframe> in your own app or dashboard |
The viewport is locked at 1440×900 — the resolution most computer-use models are trained on — for maximum grounding accuracy. Agent-opened tabs never steal focus from the user's browser.
Embedding
<iframe
src="https://app.molar.it/live/run_abc123?mode=readonly"
width="1440" height="900" allow="clipboard-read">
</iframe>
For agents
molar_run_task, molar_verify, and molar_explore_target return live-view information for browser-backed work. An agent working in a chat client can hand the returned URL to its user so they can watch the work in progress.
Recording
Runs are recorded automatically — no flag required.
| Artifact | Format | Use |
|---|---|---|
| Session video | WebM / MP4 | Share a failure, attach to a bug |
| Viewport + DOM timeline | Trace | Scrub, inspect elements at any step |
| HLS replay stream | .m3u8 | Embed a replay in the dashboard or an external app |
| Network log | HAR | Inspect every request/response |
Recordings are retained per your plan's retention window and are encrypted at rest — see Trace security.
Session inspector
Open any recorded run in Trace to get the full inspector:
- Timeline — every step with before/after screenshots (Trace detail)
- Network — requests, responses, timings (debugger & replay)
- Console — logs and errors captured during the run
- AI decisions — what the agent saw, what it decided, and why, at each step
Related
- Trace — the replay and debugging surface
- Trace detail — timeline and viewport replay
- Workers & ingestion — how capture and live streaming work
- Universal automation — the runs being streamed