Sessions

Open Sessions in the sidebar to browse recorded conversations. Select a repository or project in Scope to narrow the list.

Find a session

Use All, Active, or Ended to choose a status, and the owner selector to show your sessions or another person’s visible sessions. Load more ended sessions to browse older work. The sidebar’s Search opens the command palette.

Session cards show the title, owner, harness, activity, and available usage and evaluation information. Continuations are grouped so work that spans a resume or compaction stays connected.

Open a card to view the session at /sessions/<slug>. Session detail is its own page, so you can bookmark it or send its URL to a teammate. Sharing a URL does not change access permissions.

After a server update or a burst of activity, a Catching up on recent activity or Rebuilding data views after an update notice can appear above the list. Until it clears, the list and its counts may be incomplete: a missing session is not there yet, rather than gone.

Session detail

The default tab is Overview. Four tabs answer different questions:

Tab	Use it for
Overview	Outcome, tasks, plan artifacts, metadata, and related work.
Transcript	The conversation, organized into turns.
Events	The underlying recorded events and tool activity.
Evaluation	Scores, findings, evidence, and retrospective suggestions.

A live session can have tasks and a transcript before it has an outcome. Outcomes are generated after the session ends when a summary model is available.

Plans and design documents the agent declares appear on Overview while the session is still running, and update as the agent writes them. The task list shows eight rows; expand it for the rest. A Declared chip marks a list the agent recorded, and From tool calls one reconstructed from its task tool activity. When part of a plan came from a session you cannot see, the card says its count may be incomplete.

The side panel lists every pull request the session opened, each with its state (open, draft, merged, or closed) and branch. Review on an unfinished PR starts a PR review of that PR. Tokens breaks usage down per model, with a bar splitting fresh tokens (input and output) from cached ones (cache reads and writes) and an estimated cost.

Older links to Trace or Details lead to the corresponding Transcript or Overview view.

Read a transcript

The transcript reader has a turn index and a reading pane. Select a turn to read its user prompt, assistant response, and tool calls. Filter the index to find a relevant exchange, or use the previous and next controls.

Choose Full to inspect tool activity, or Summary for a shorter reading view where a summary exists. Expand tool results when you need the evidence behind a response. Older turn details load when you open them.

Each subagent appears as its own row under the turn that spawned it. Select the row to read the subagent’s messages in place, in the same reader; filtering the index by an agent’s name finds it. A background command reports its outcome on the call that started it rather than as an agent row.

Shell commands show their output as output, in one card with the command, with column alignment kept. The reader keeps your place when the page reloads.

For live sessions, the reader follows the newest turn until you select another one. This lets you inspect an earlier exchange while work continues.

Token and cost displays use the usage reported by the harness. Cache reads and writes are accounted for separately where available; estimated cost depends on known model prices.

Inspect events

Open Events when the conversation alone does not explain what happened. Recorded tool calls, results, errors, and lifecycle events let you inspect the sequence behind the session.

What appears depends on what the harness exposes and what kcap captured. A session recorded through an import can differ from one captured live.

Review the outcome and quality

Session outcomes explain what was accomplished and show the session’s plans and tasks. Evaluations assess the work and provide evidence for their findings.

Use Session recap and recall to bring earlier work into your agent conversation or read a recap from the terminal.

Control access and recording

Use the session’s sharing controls to adjust who can see it. Visibility & sharing also covers making a session private, stopping capture, and excluding repositories.