Neuro OS: agent output becomes first-class

Neuro OS makes what your agents produce first-class: every work product visible on one page, video playable in the thread, decisions returned as structured data, and wireframes your team can review before a build. Each of those has its own deeper note; this is the catch-up.

A company OS is not measured by how many sessions you opened. It is measured by what landed — files, recordings, confirmations — and whether a human can review them without reconstructing the day from logs.

Every work product, one page

A project-scoped Artifacts page indexes every file, video, and document your agents produce across sessions and change requests, grouped by task stack by default. Reviewing a day of output no longer means opening each session one by one.

Artifacts sit next to the git-backed operating layer: the same company that versions agents and skills can browse what those agents shipped. Nothing to turn on — once sessions produce files, they show up.

Video plays in the thread

Sessions and issues accept video attachments and render rich inline previews, player controls included — so a screen recording lands in the conversation as something you press play on, rather than an opaque download. Playback works where the team already works: web, and the same surfaces you use for review.

That matters for handoffs. An agent that recorded a broken flow can attach the clip; the next human (or the next agent) sees the evidence without hunting a Drive folder.

Answers come back structured

An agent can ask the board or a user to confirm one or more options through a checkbox card in the session thread — validated the same way across API, CLI, and UI. The response is data your automation can act on, not prose to parse.

That is the same philosophy as policy on connectors: the control plane wants a typed decision, not a hope that the model “understood” a paragraph. Confirmation cards close the loop between agent proposal and human gate without inventing a second review system.

Wireframes before the build

The release also bundles a wireframe skill. Point an agent at an app idea and it returns a reviewable deck: one SVG per screen, desktop and mobile variants, a flow map, and an HTML viewer. Install it from the skills catalog; the skill itself is files in your repo — purpose, steps, house style — not a hidden prompt.

Review the deck on a change request while changing the answer is still cheap. Then build.

Harder-working agents

Operators get an audited clear-error action and live-session stop finalization. Invalid agents are centrally blocked from receiving new assignments. Document-scoped comments read as review context instead of waking the assignee like a top-level ping.

Your instance tries harder before involving you, and wakes agents on the activity that actually needs them — the same “more capability, not more authority” instinct as AGI-ready architecture.

Also in this wave: a collapsible sidebar rail, an opt-in refresh of project and agent surfaces, automated PR quality and security gates, and a round of reliability fixes. For team-scale credentials and timelines, see Run Neuro OS with your whole team and user-scoped secrets.

Review the work itself

Together these change what a morning review looks like: open Artifacts, play the recordings, mark up the wireframes, and answer the confirmation cards — without reconstructing any of it from session logs.

The operating layer stays files you own. The outputs become first-class citizens of that layer. That is how a workforce compounds instead of resetting every chat.

Open Artifacts. Review what landed.

Start on your project’s Artifacts page, install the wireframe skill, and treat agent output as something the company can see. Explore Neuro OS.

Explore Neuro OS →