Model routers & agent gateways - DataVibe Documentation

A router or agent gateway selects a model and returns completion text. DataVibe then evaluates the completed send-bound payload as custom policy middleware:

Application → agent gateway / model router → completed output → DataVibe gate → CRM / sequencer

This is the post-generation slot — complementary to platform safety (e.g. Model Armor on Vertex), not a substitute. Normalize router output to recipient, subject, body_text, body_html, and source_model, then call POST /v1/gate/outbound.

{
  "recipient": "prospect@example.com",
  "subject": "Following up",
  "body_text": "Hi Jordan, thanks for the conversation.",
  "body_html": "<p>Hi Jordan, thanks for the conversation.</p>",
  "source_model": "router/selected-model",
  "idempotency_key": "send-abc123"
}

Reference pack (LiteLLM callback, Portkey hook, runnable demo):

Monorepo: integrations/agent-gateway/
Partner one-pager: docs/partners/AGENT_GATEWAY_POLICY_MIDDLEWARE.md
Public docs: Agent gateway policy middleware

The production integration belongs after routing because DataVibe must evaluate the exact content that would reach the customer. Keep router provider keys in your own infrastructure.

Use the persisted gate for outbound. POST /v1/check is a stateless preview and cannot create a queue item or review URL.

Last modified on September 14, 2026