Backstage

Live layer for conferences

The live layer for everyone who is not in the room.

A conference runs eight rooms at once and narrates itself in public while it does. Backstage reads that record as it is written, places every post on the session it is about, and keeps the result in a graph: which room is loud right now, what just shipped that touches your code, which ten minutes you missed.

The tour follows one real post end to end. The console replays GitHub Universe 2025 minute by minute and goes live for Universe 2026 on 28 October.

Moment · 10:06 PT · Main Stage
What the venue is posting about

“A mission control to assign, steer, and track Copilot coding agent tasks” is landing

    author is the speaker · posted during the slot6 posts
    151

    sessions on the agenda

    RainFocus catalog, GitHub Universe 2025
    143

    posts placed on a specific session

    of 286 on the record, each timed from its post ID
    718

    overlapping session pairs

    the conflict graph, computed in FalkorDB
    33

    announcements traced to stacks

    github.blog and the changelog, 27 to 31 Oct

    The problem

    Most people who care about a conference are not in the room. The ones who are, are in the wrong one.

    Organizers are cutting remote access because livestreaming costs too much. Research conferences are opening second sites because speakers cannot get visas. Recordings land weeks later with no index. Inside the venue it is the same problem at a smaller scale: at any minute, every attendee is outside seven of eight rooms, and the day is being narrated on Bluesky, LinkedIn and X by the people who are there.

    “Costs for supporting livestreaming at the conference venue have simply become cost-prohibitive.”CSCW 2025 organizers, on cutting remote attendance
    “The number one piece of feedback after the EU event was that our schedule builder left something to be desired.”CNCF, after KubeCon Europe

    How it works

    The public record, resolved to the agenda, with a reason on every edge.

    Four agents share one FalkorDB graph. The Listener ingests posts and reads their exact time out of the post ID. The Resolver decides which session each post is about and writes the evidence onto the edge. The Wire tags every announcement with the parts of a stack it touches. The Curator turns bursts into Moments. Every screen is a query against that graph, and every answer comes with its path.

    01 Public recordBluesky · LinkedIn · XExa discovery · Bright Data for the hard parts
    02 Listenerexact to the milliseconddecoded from post IDs
    03 Resolverpost → sessionauthor · title · time · stage
    04 GraphFalkorDBpaths, not lookups
    05 Youmoments · wire · your stackand a line back from the room

    Product

    Three things you get that no conference app gives you.

    Moments

    The minute you would have walked over for

    A spike in one room, an announcement starting to land, the whole venue posting at once. Each Moment carries its evidence and its time, so you can walk over now or jump to those minutes in the recording later.

    Wire

    Announcements matched to your repository

    Every changelog entry and launch post is tagged with the parts of a stack it touches. Paste a repository and the graph answers which of today's announcements change something you depend on, and which sessions are yours.

    From the room

    One line back from inside

    Anyone in a room can send a single sentence without an account. It is placed and attributed like any post and reaches every open console the moment it lands. That makes the layer two-sided.

    Built on

    A graph that can explain itself, fed by data a plain script cannot reach.

    FalkorDB · graph and agent memoryBright Data · SERP, Unlocker, Browser APIExa · discoveryOpenAI · embeddings and adjudicationFastAPI · server-sent eventsFly.io