The standup runs itself.

Scriber facilitates your standup by voice — prompting each person, capturing blockers, updating Linear in real time. A second silent agent, Mnemo, remembers every prior standup and whispers what you forgot.

Built in one night for OpenAI Voice Hack Night, Cerebral Valley.

Why this is different

Every meeting AI is passive. Scriber acts.

Spinach, Otter, Fireflies, Fellow, MeetGeek — they listen, then post-meeting they act. Scriber flips this. The agent acts during the conversation, with the Linear board visibly mutating live. The visceral moment: someone describes a flow, Scriber generates a diagram with gpt-image-1, and the image appears on the ticket in the Linear UI before the speaker finishes their sentence.

Facilitator

Scriber

Joins your call. Prompts each teammate. Listens for blockers, commitments, status changes. Executes Linear actions as the conversation happens.

Voice
cedar
Model
gpt-realtime-2
Reasoning
medium · low latency
Supervisor

Mnemo

Never speaks into the call. Reads the live transcript. Holds context across every prior standup. Whispers tactical guidance to Scriber over a private side-channel.

Voice
silent · text only
Model
gpt-realtime-2
Reasoning
xhigh · deep thinking
How it works

Two agents. One conversation. Real action.

  1. 01

    Scriber joins your standup

    Voice in, voice out — over Google Meet or Zoom. Speaks like a colleague, not a bot.

  2. 02

    It executes Linear actions live

    Creates tickets. Updates statuses. Adds comments. Generates diagrams via gpt-image-1 and attaches them to tickets — all mid-sentence.

  3. 03

    Mnemo whispers what Scriber would miss

    "Vishal mentioned this latency issue Tuesday — Ben committed to spiking it." Scriber voices it naturally.

  4. 04

    The standup wraps itself

    Verbal recap. Slack message. Linear board reflects everything that was said. Total human effort: just talking.

Demo

Three minutes. One real standup.

REC

Demo video drops at submission — refresh in a few hours.

Architecture

Two Realtime sessions. One WebSocket. Browser-native.

                  ┌────────────────────────────┐
                  │  Google Meet / Zoom call   │
                  │  humans + Scriber          │
                  └──────────┬─────────────────┘
                             │  audio in/out (BlackHole 2ch)
                  ┌──────────┴────────────────────┐
                  │   Browser (Next.js page)       │
                  │                                │
                  │  ┌──────────────────────────┐  │
                  │  │  Scriber                 │  │
                  │  │  Realtime-2 · cedar      │  │
                  │  │  Tools: linear.* ·       │  │
                  │  │   diagram · slack        │  │
                  │  └──────────────────────────┘  │
                  │             │       ▲          │
                  │   transcript│       │ whispers │
                  │             ▼       │          │
                  │  ┌──────────────────────────┐  │
                  │  │  Wake-word state machine │  │
                  │  │  ALWAYS_ON → LISTEN →    │  │
                  │  │  ACTIVE → QUIET → prev   │  │
                  │  └──────────────────────────┘  │
                  └──────────┬─────────────────────┘
                             │ WS
                             ▼
                  ┌────────────────────────────────┐
                  │  Mnemo (Bun sidecar)           │
                  │  Realtime-2 · text-only        │
                  │  Tools: linear.read · rag ·    │
                  │   whisper_to_scriber           │
                  └────────────────────────────────┘

Tonight's build uses BlackHole 2ch for audio routing on macOS — Scriber lives in a browser tab, audio bridged into Meet through a virtual cable. Production swaps to Recall.ai so Scriber appears as a real tile in the Meet grid.

Stack

What's under the hood

Want this for your team?

Drop a note — I'll wire you up when the hosted version ships.

Email the waitlist