We’ve built a Java service to push Agent Assist keyword additions through the /api/v2/agent-assist/keyword-sets/{id}/keywords endpoint, but the atomic POST keeps failing with a 400 Bad Request when the synonym matrix exceeds three entries. The payload constructs keyword ID references, sets case sensitivity to false, and runs a duplicate check against a local Redis cache before hitting the assist engine. Here’s the minimal repro for the addition request: {“keyword”: “billing inquiry”, “case_sensitive”: false, “synonyms”: [“bill question”, “payment help”, “charge issue”], “metadata”: {“audit_log”: true, “rebuild_index”: true}}. The validation pipeline strips special characters and verifies the maximum keyword count limit, yet the response body returns {“code”: “invalid_request”, “message”: “Keyword addition violates assist engine constraints.”}. Latency tracking shows the request hangs at 2.4 seconds before the timeout, which suggests the automatic index rebuild trigger isn’t completing.
Can you clarify if the assist engine requires a separate format verification step before the atomic POST, or should the webhook callback handle the synchronization with our external knowledge base API? We don’t want the addition to silently fail during scaling. The audit logs aren’t capturing the index update rate, so we’ve got no visibility into why the pipeline drops the payload after the special character verification step. Pretty sure the atomic operation is blocking.