Everything it does, on every surface
Inline in the editor, across the panels, in the terminal, and in both editors — every surface OAK gives you, at a glance.
Every agent and its changes live in one master-detail panel — pick a session, agent, or prompt, and its change map fills the right.
Every session on the workspace's machine, grouped by workspace — its title, model, tokens and store size. Pick one and every panel follows.
Pending edits appear in the editor itself — a tinted line, a gutter star, and a lens with Keep, Undo, the diff, and the way to the full review.
The conversation as it happened — your prompts, the agent's replies and thinking, and every tool call with the diff it produced.
The session reads as the conversation you had, one row per ask. Select one to scope everything beside it.
Observations pair the change feed with the agent's own reasoning, parsed from the transcript at zero tokens.
Every edit gets its own diff, stacked into one scroll — each with its own Keep and Undo, not one session-wide tangle.
The review scoreboard — pending, accepted, reverted — sits over token and usage plots.
Spotlight dims every line the agent didn't touch, so the changes read at full contrast.
Edit the same lines and the undo stops — surfacing the clash instead of overwriting your work.
Two audits cover what ran that could harm and everywhere the session reached off-machine.
One click builds a context-preloaded prompt about an edit — a draft on the clipboard, sent to the session's agent only when you choose.
herdr, the Observatory and Review in one terminal — your agents in herdr panes beside a live view of every session.
One backend serves every surface — and every view is a --json command.
oak demo replays a scripted session through the real capture pipeline — every panel fills in live.
See each surface in depth in the docs.