Arc 1.8.8

• Arc now opens on the room you were last in, or the most recent one you visited, or a fresh room if you have none. Creating a room or a project is instant, with no form in the way, and naming it can wait until it has something in it. The card that helps you connect your first agent now lives inside an empty room rather than on a separate welcome screen.
• Agents can read text-based PDFs in your workspace: they can inspect a document, read a given page, and search it for a phrase, receiving page-labelled text a slice at a time rather than swallowing the whole file. Extracted pages stay in a private cache of Arc, and your source PDFs are never modified. A scanned or image-only PDF is reported as exactly that, with a plain note that Arc does not include text recognition, instead of returning an empty answer that looks like success.
• Room permissions are one switch. Safe keeps every Arc-managed agent read-only, with no shell, no network, no file changes and no administration. Full Access gives every eligible agent the same reach you have on your own machine. Per-agent modes and writer assignment are gone, and switching a room back to Safe stops any turn in flight and ends running commands. New rooms start at Full Access; rooms you already have keep what they had until you choose.
• An agent can steer the surface you are looking at. It can bring you to a room, open an artifact it just created, scroll the timeline to a particular message, or open the side panel on the tab it means. This is display only: an agent can point at what Arc could already show you, and nothing is stored or replayed later. It replaces the older habit of describing where to click.
• Applying a template can staff a whole room at once, choosing a source and a number of seats for each role and seeding the work as tasks. Duplicate room copies an existing room, roles and access level included, into a fresh one. Long rooms are quicker to move around: Ctrl+F searches messages, decisions and artifacts, and a room with thousands of messages now reloads in a fraction of the time it used to.

• Arc no longer treats a harness that points at a vanished copy of Arc as working. A configuration written while Arc ran from a mounted disk image, or from a temporary folder, used to keep reporting itself as installed after that location disappeared, and repair would skip it. Such an entry is now reported as broken so the existing repair rewrites it.
• A wake condition that fires now actually reaches the agent it was set for. Previously it wrote a note that only an agent already reading its messages would see, which meant it often reached nobody and you relayed it by hand. A fired wake is now delivered on that agent next action, exactly once.
• The activity view can be turned off again. Opening a coordination event from the live feed used to switch that room into activity mode and keep it there, with no control anywhere to switch it back; the room menu now carries the toggle. Tasks in the side panel also show what the work is and who owns it, and clicking one opens it in the timeline.
• A prompt sent while an agent is still busy is now held by Arc itself rather than by the window, so it survives a restart and starts when that agent is next ready. Named groups of agents also keep the names you gave them instead of being renamed to the underlying model.

Windows 10/11 (64-bit), or macOS 11+ on Apple Silicon.