𝗕𝘂𝗶𝗹𝗱𝗶𝗻𝗴 𝗮 𝗦𝗲𝗿𝘃𝗲𝗿𝗹𝗲𝘀𝘀 𝗪𝗲𝗯 𝗦𝗲𝗮𝗿𝗰𝗵 𝗦𝗲𝗿𝘃𝗶𝗰𝗲 𝗳𝗼𝗿 𝗔𝗜 𝗔𝗴𝗲𝗻𝘁𝘀
AI agents need web search to find current facts.
When many teams build agents, they often use different search providers. This creates a mess. You end up with different APIs, different costs, and fragmented tools.
At PostNL, we solved this by building a centralized search service on AWS.
We built a system that works like this:
• A single interface for all teams. • Support for many search backends. • Low operational work. • Low costs. • The ability to change providers without breaking anything.
Our architecture uses a routing layer. This layer sits between your AI application and the search provider.
The core parts of the system:
- API Gateway: Provides a private entry point for authorized users.
- AWS Lambda: Runs a lightweight router written in Go.
- ECS Fargate: Hosts our initial search backend using SearXNG.
We chose Go for the router because it is fast and uses little memory. It works perfectly as a small Lambda function. This setup scales automatically as you use it more.
We also used a hexagonal architecture. This means the core logic does not care which search engine you use. Today we use SearXNG. Tomorrow we can switch to a commercial provider by simply adding a new adapter. The users will never notice a change.
The goal is simple. We provide the search data. The AI agent handles the reasoning and the answers.
By centralizing this service, teams stop worrying about search APIs and start focusing on building better AI solutions.
Optional learning community: https://t.me/GyaanSetuAi