Self-Hosted Trieve RAG Platform on AKS

Overview

The platform needed retrieval-augmented generation without sending proprietary client data to third-party vector APIs. I self-hosted the entire Trieve stack on AKS.

The problem

The platform needed retrieval-augmented generation without sending proprietary client data to third-party vector APIs.

What I did

1. 1

   Self-hosted the full Trieve stack on AKS: Qdrant for vector storage, embedding and reranker models, Keycloak for SSO, and ClickHouse for analytics.

2. 2

   Integrated the stack with browser-use AI agents already running on the platform.

3. 3

   Brought the deployment under the same observability and alerting used for the rest of the platform.

Tech used

Outcomes

• In-house retrieval pipeline with zero external data egress.

• Fully integrated with the platform's existing observability and alerting.