Builders Fredrik Björk and Julius de Bruijn have launched an open supply challenge for synthetic intelligence routing — choosing the right mannequin for the precise job, whereas additionally serving as an aggregator for a number of Mannequin Context Protocol (MCP) servers: Nexus.
“As AI purposes grow to be extra subtle, they more and more must work together with a number of exterior companies, APIs, and totally different language fashions,” the pair explains of the issue the Nexus challenge got down to resolve. “This creates a number of ache factors: context, helps the LLM choose from a doubtlessly giant variety of MCP instruments with out overwhelming it; price, lack of strategic mannequin choice can result in pointless bills; observability, supplies insights into the efficiency and conduct of the AI system, enabling higher decision-making and troubleshooting; [and] safety, ensures that requests are routed securely and in compliance with governance insurance policies.”
Nexus guarantees to ship an open supply strategy to routing requests to totally different LLMs and MCP servers. (📷: Nexus)
Nexus seems to be to repair these issues by providing automated routing of queries to totally different giant language fashions (LLMs) — bearing in mind totally different fashions’ efficiency at totally different duties, the latency between immediate and response, context window lengths, and availability. This, the pair say, might be enhanced sooner or later to incorporate extra superior routing algorithms, help for customized guidelines and insurance policies, and real-time monitoring.
The software program additionally serves a second objective: appearing as a proxy between “agentic” AI and the Mannequin Context Protocol (MCP) servers which give them their capabilities. A single request to Nexus, the builders say, might be robotically routed to probably the most acceptable MCP server for the duty – full with authentication and connection administration, and the power to combination responses from a number of sources if required.
“The way forward for AI purposes lies in clever orchestration,” Björk and de Bruijn declare, “and Nexus is our contribution to creating that future extra accessible and environment friendly for builders in all places.”
Nexus is accessible, together with its supply code, on GitHub underneath the Mozilla Public License Model 2.0; extra data is accessible on the challenge web site.
