Skip to main content

Documentation Index

Fetch the complete documentation index at: https://docs.mycelium-ai.co/llms.txt

Use this file to discover all available pages before exploring further.

WhatsApp uses an in-process bridge rather than an HTTP webhook receiver. The bridge connects to the WhatsApp multi-device protocol and exports to the vault on a configurable cadence.

​
Open-source ingest skill

Pull recent messages from a named WhatsApp chat into External Inputs/WhatsApp/<contact>/<YYYY-MM-DD>.md. Requires the WhatsApp bridge running locally. 
/ingest-whatsapp "Nelly Ortiz" --days 7
Voice notes are transcribed by the bridge before export. Each voice note renders inline as [Voice note] <transcript>.

​
Bridge setup (paid runtime)

The bridge is a Go binary that connects to the WhatsApp Web multi-device API. It runs as a sidecar to the runtime, not as an HTTP endpoint.

​
Prerequisites

  • WhatsApp account with multi-device enabled.
  • Bridge binary deployed alongside the runtime container.

​
Configuration

# bridge.yaml
bridge:
  vault_export_path: vaults/{tenant_id}/External Inputs/WhatsApp/
  transcription: whisper-base  # or whisper-small for higher accuracy
  export_interval_hours: 4
  pending_transcript_timeout_seconds: 180

​
What gets exported

Content typeMemory categoryNotes
Text messagefactSender, timestamp, thread context
Voice notefactTranscript inline; bare [Voice note] if expired
Shared documentfactFile name and type; content not ingested

​
Privacy

Voice note audio is never stored in the vault. Only the transcript lands in the markdown file. Audio files are processed in memory and discarded.

​
Retry

Voice note transcription retries automatically on failure. Expired audio (deleted from WhatsApp’s servers before bridge captured it) is logged as orphan_no_keys and marked as a bare [Voice note] with no transcript.