feat(vapi): add Vapi voice AI webhook memory integration

[benfrank241]

Adds Hindsight persistent memory to Vapi voice AI calls via a webhook handler.

How it works

Unlike Pipecat (per-turn frame processor), Vapi injects memory once per call via webhooks:

• assistant-request — recalls memories for the caller (keyed to phone number) and returns assistantOverrides with a system message containing those memories
• end-of-call-report — retains the full call transcript asynchronously (fire-and-forget, never delays the webhook response)
• All other event types — returns None (HTTP 200 no-op)

Outbound call helper: build_assistant_overrides(query) for injecting memory at call creation time.

Usage

from fastapi import FastAPI, Request
from hindsight_vapi import HindsightVapiWebhook

app = FastAPI()
memory = HindsightVapiWebhook(bank_id="user-123", hindsight_api_url="...")

@app.post("/webhook")
async def vapi_webhook(request: Request):
    event = await request.json()
    response = await memory.handle(event)
    return response or {}

Changes

• hindsight-integrations/vapi/ — new hindsight-vapi package
• .github/workflows/test.yml — CI job + detect-changes
• scripts/release-integration.sh — added "vapi" to VALID_INTEGRATIONS

Tests

20 passing unit tests covering: client resolution, assistant-request recall+inject, end-of-call retain, outbound overrides helper, unknown event no-op.