Docs/Integrations/Integrations

    WhatsApp Groups

    Last updated · MAR 2026·Read as Markdown

    WhatsApp Groups integration lets your AVCodex agent learn from real conversations happening in your WhatsApp groups. Instead of manually creating knowledge sources, the agent's knowledge base stays current by automatically ingesting group chat transcripts through the RAG pipeline.

    This is separate from 1:1 WhatsApp messaging where users chat directly with your agent. Group integration is about listening and learning, not responding.

    Note: WhatsApp Groups requires an active WhatsApp Business connection. Set up WhatsApp integration first.
    1. You connect a WhatsApp group (for example, your "Field Tech Daily" group) to your agent by providing the group ID
    2. Incoming group messages are buffered in the database (text messages only)
    3. On a configurable schedule, buffered messages are formatted into a conversation transcript
    4. The transcript is processed through the RAG pipeline, chunked, embedded, and stored as a knowledge source
    5. Your agent can now reference the group's conversations when answering questions

    Each connected group becomes its own knowledge source, labeled "WhatsApp Group: [name]" in your knowledge sources list. The knowledge source is updated incrementally as new messages arrive and get ingested.

    Each group can be configured with a mining mode that controls how messages are used:

    Mode Behavior
    Knowledge Messages are buffered and periodically ingested into the RAG knowledge base. The agent learns from conversations but does not participate.
    Reactive Messages are available for real-time context but not permanently stored in the knowledge base.
    Both Combines knowledge ingestion with reactive access.

    The default mode is Knowledge, which is the most common use case.

    1. Ensure WhatsApp Is Connected

    Your agent must have an active WhatsApp Business connection. Go to Share and verify the WhatsApp deployment is configured and active.

    2. Get the Group ID

    The group ID is provided by the WhatsApp Business API when your number is added to a group. You can find it in the webhook payloads or through the Meta Business Platform.

    3. Add the Group

    Use the API to register the group for mining:

    `` POST /api/applications/:appId/whatsapp-groups ``

    ``json { "groupId": "120363012345678901@g.us", "groupName": "Field Tech Daily", "miningMode": "knowledge", "ingestionIntervalMinutes": 1440 } ``

    4. Wait for Messages

    As group members send messages, they are automatically buffered. At each ingestion interval, the buffered messages are processed into your knowledge base.

    Field Range Default Description
    groupName -- -- Friendly name for the group (used in the knowledge source label)
    miningMode knowledge, reactive, both knowledge How messages are processed
    ingestionIntervalMinutes 60-10,080 1,440 (24h) How often buffered messages are ingested into RAG
    isActive true/false true Pause or resume message buffering

    Ingestion Interval#

    The interval controls how frequently the cron job processes buffered messages into the knowledge base. Shorter intervals mean more current knowledge but more processing. Longer intervals batch more messages into each ingestion cycle.

    Interval Good For
    60 min (1 hour) Fast-moving groups during a major commissioning week
    1,440 min (24 hours) Most groups, daily knowledge updates
    10,080 min (7 days) Low-traffic groups (a quiet dealer-program chat) or cost-sensitive setups

    Only text messages are buffered and ingested. Images, videos, documents, stickers, and other media types are silently skipped. (A photo of a fault LED is not ingested. The text reply that says "that's the Tesira AVB grandmaster mismatch" is.)

    Messages are formatted as a timestamped conversation transcript before being processed through the RAG pipeline:

    code
    WhatsApp Group: Field Tech Daily
    Messages: 47
    
    [2026-03-23 10:15:00 UTC] Sarah (+1 212 555 1234): Boardroom 4's DM-NVX-360 is dropping audio every 12 min, anyone else?
    [2026-03-23 10:17:00 UTC] Mike (+1 310 555 5678): @Sarah we saw that on a 1.6111 firmware build. Roll back to 1.6010.
    [2026-03-23 10:18:00 UTC] Sarah (+1 212 555 1234): Confirmed, that fixed it. Thanks.

    Each ingestion cycle creates or updates the group's knowledge source with the new transcript, which is then chunked and embedded for retrieval.

    List Configured Groups#

    code
    GET /api/applications/:appId/whatsapp-groups

    Returns all group configs with message statistics (total messages, pending messages, last message timestamp).

    Update a Group Config#

    code
    PUT /api/applications/:appId/whatsapp-groups/:id
    json
    {
      "miningMode": "both",
      "ingestionIntervalMinutes": 360
    }

    Pause a Group#

    Set isActive to false to stop buffering new messages without deleting the config:

    json
    {
      "isActive": false
    }

    Trigger Manual Ingestion#

    Instead of waiting for the next scheduled ingestion, trigger it immediately:

    code
    POST /api/applications/:appId/whatsapp-groups/:id/ingest

    Returns the number of messages processed and chunks created.

    Remove a Group#

    code
    DELETE /api/applications/:appId/whatsapp-groups/:id

    This removes the group config and its buffered messages. The knowledge source created from previous ingestions remains in your knowledge base until you delete it separately.

    Use Case Example
    Internal knowledge capture Mine your "Field Tech Daily" group so the agent learns every fix the team discovers in real time
    Dealer program FAQ Learn from a dealer-support group to answer common dealer questions automatically
    Manufacturer firmware tracking Track manufacturer-rep groups (Crestron, QSC, Biamp) to keep the agent current on firmware quirks
    Bid intelligence Mine the bid-team group for RFP context the BD agent can reference
    Commissioning escalation Capture commissioning lead discussions so the field-facing agent learns from real resolutions

    Messages are deduplicated by WhatsApp message ID per group config. If the same message arrives twice (webhook retries), only one copy is stored. This prevents duplicate content in your knowledge base.

    No messages being buffered?

    • Verify the group ID matches exactly (including the @g.us suffix)
    • Check that the group config isActive is true
    • Confirm your WhatsApp Business number is a member of the group
    • Only text messages are buffered, media messages are skipped

    Knowledge source not updating?

    • Check the ingestion interval, the cron job only runs when the interval has elapsed
    • Use the manual ingest endpoint to trigger immediately
    • Verify the mining mode is set to knowledge or both

    Duplicate group error (409)?

    • Each group ID can only be configured once per agent
    • Delete the existing config first if you want to reconfigure
    Tip: Combine WhatsApp group mining with scheduled Heartbeat messages to build an agent that proactively shares insights from the field-tech group with the dispatch team.

    *AVCodex · Your AV expertise. Amplified by AI.*

    Was this helpful?
    Edit this page →