Pular para o conteúdo
TakoAPI
Voltar às skills

percept-listen

por jarvis563

Coding Agents & IDEs
261Downloads informados pelo clawskills.sh
0Visualizações
0 curtidaEntre para curtir esta skill
—Ainda sem avaliações
Sua avaliação

Resumo

Ambient audio capture and transcription for OpenClaw agents via wearable devices.

Descrição

O que esta skill faz

Connects a wearable microphone (Omi pendant or Apple Watch) to an OpenClaw agent for ambient audio capture. Audio is transcribed on-device, then streamed as speaker-tagged, timestamped conversation data to a local Percept server. All processing stays local — no audio is sent to external services.

All transcription and storage runs locally via SQLite with full-text search, so no audio or transcript data leaves the user's machine.

Quando usá-la

  • Logging meeting notes hands-free during in-person discussions
  • Capturing ambient conversations for later review or search
  • Triggering agent actions based on spoken wake words
  • Searching past conversations with full-text lookup
  • Recording field notes while keeping hands free

Exemplo de fluxo de trabalho

ENTRADA

User says: 'Start listening to my meeting'

AGENTE
  1. 1

    Agent confirms Percept server is running on port 8900

  2. 2

    Omi pendant captures audio and phone performs speech-to-text

  3. 3

    Transcript segments arrive via webhook and Percept processes them into a conversation

  4. 4

    Conversation is stored in local SQLite DB with speaker tags and timestamps

  5. 5

    Agent surfaces a searchable summary after the session ends

SAÍDA

Timestamped, speaker-tagged conversation stored locally and queryable via full-text search

Requisitos

  • Omi pendant paired via the Omi phone app, OR Apple Watch with Percept app installed
  • Percept app webhook configured: Omi app → Settings → Webhooks → your tunnel URL
  • Tunnel service account (Cloudflare Tunnel, ngrok, or Tailscale) to expose local server

Instalação

clawhub install jarvis563/percept-listen

Execute o comando da CLI ou cole o prompt no Claude Code, Cursor, Codex ou no agente de código que você usa.

Links