Activity: runs and sessions.

The two tabs

Activity, in the rail, has two tabs. Runs holds the recordings of what has already run. Live shows the sessions your paired sidecar is holding open right now. Neither needs a sign-in.

Runs

Recordings of agent, workflow and strategy runs, plus the Recovery log.

Live

The machines a paired daemon is running for you, each with its own controls.

What is recorded

Agent, workflow and strategy runs are recorded as they happen. An agent recording holds the prompts, the replies, each tool call with its arguments, each tool result with how long it took and whether it failed, and any artifacts the run produced.

Recordings are kept in this browser, separately for each signed-in identity. With a workspace connected, each run is also written into the workspace and read back from there, so your history survives a cleared browser.

Saved, Recent and Deployed

The Recordings list is grouped into Saved, Recent and Deployed, each with a count. A row shows the run’s title, how long ago it ran, and a failed mark if it failed.

Save

Keeps a run. Saved runs are never trimmed; Unpin returns one to Recent.

Recent

The last 100 unsaved runs. When a new one arrives, the oldest unsaved run goes first.

Clear recent

Empties Recent and leaves Saved alone.

Delete

Removes one recording.

Replay a run

1. Select a recording. It opens on the finished transcript.
2. Press Play to watch it unfold step by step. Pressing Play at the end starts again from the beginning.
3. Click or drag the track to jump to any point. Grabbing the track pauses playback.
4. The speed button cycles through 1×, 2× and 4×.

Coloured marks on the track show each prompt, reply, tool call, artifact, error and the end of the run. Hover a mark to see what it is.

Recovery

The Recovery tab is the self-heal log: what failed in the middle of a run and what recovered it. It totals each source as recovered, exhausted or abandoned, links each entry to the nearest recording, and has a Clear button.

Playback in chat

The same playback track can sit inside a live conversation, a crew run and Craft, so you can scrub back through a run before it ends. It is hidden by default: turn it on in Settings under Playback in chat.

While the run is going the track shows ● live. Scrub back to see the run as it was at that point, and Jump to live returns you to the latest step.

What a session is

• A paired sidecar, started with --daemon. See Running the sidecar.

A session is a machine the daemon holds open for work. Each card says whether it is Isolated, in its own microVM, or On the host, running directly on the sidecar’s machine. The badge reads what the daemon reports. If any session is on the host, the page warns you that nothing confines what it can reach.

Sessions run on the host unless you start the sidecar with microVM isolation, which needs the msb runtime installed. On Windows that runs inside WSL2.

Start one

1. On the Live tab, choose New session.
2. Give it a working directory on the sidecar’s machine. It is mounted into the session read-write.
3. Choose Start.

If the daemon was started with --clone-under, you can start from a repository instead, with a branch, tag or commit. The sidecar clones it with its own git credentials, which never enter the session, and nothing is pushed back. The card then shows the repository, the ref and the commit.

Watch, pause, capture, destroy

Watch

Streams the session’s CPU, memory and output to this page. Nothing streams until you press it; Stop watching ends it.

Pause / Resume

Pausing tells you whether any calls were still running and did not finish.

Capture

Saves the paused machine as a snapshot. Only microVM sessions can be captured, and only once paused.

Destroy

Asks first, with Destroy or Keep, then removes the session.

A new session can start from a capture: pick it under Start from instead of A fresh machine. The capture restores the machine’s disk, and the working directory is mounted fresh on top of it.

Limits

Each microVM gets 1 CPU, 512 MB of memory and an 8 GB disk by default. All captures together are capped at 32 GiB unless you set another limit with --snapshot-budget; the page shows how much of it is used.

A capture that would go over the cap is refused. Corenel never deletes an older capture to make room.