Dev wires SearXNG web search into their AI agent as a sub-agent
A developer who has been building an AI agent app since June 2025 recently added SearXNG, an meta , after previously only having basic web capability. Seeing repeated mentions of SearXNG, they integrated it directly (with AI assistance) into their app. The setup works as follows: the (the main chat model) calls a dedicated tool called WebResearch, which spins up a sub-agent.
That sub-agent runs on a secondary GPU using a 3.5 9B model, and it only has access to two tools — WebSearch and WebFetch — using SearXNG as the . Following a set of rules and instructions, the sub-agent conducts the research and produces a markdown report that feeds back to the main chat model. This mirrors a pattern already used elsewhere in the app via a general-purpose spawn_agent tool, which spins up with different roles, but that tool is only available in the app's Workspace mode.
Since Chat mode lacks access to spawn_agent, the developer built a separate wrapper specifically so Chat mode could still safely use the WebResearch capability.
Key points
- Integrated SearXNG, an meta , via a dedicated WebResearch tool
- The calls WebResearch, which launches a sub-agent (a 3.5 9B model on a secondary GPU)
- The sub-agent only has WebSearch and WebFetch tools, follows set rules, and returns a markdown report
- A general spawn_agent tool already does similar sub-agent spawning but is Workspace-mode only, so a separate wrapper was built for Chat mode