Skip to main content
Caleb Hunter
In Development

OmniRoute

Talk to one AI. Get the best AI.

Voice-first AI meta-router that uses Elasticsearch as a living capability registry to intelligently route requests to the optimal model — or chain multiple models together for complex tasks.

AI Involvement

OmniRoute is a meta-AI system: it uses AI to decide which AI to use. An orchestrator model receives the user's voice request, then queries an Elasticsearch capability registry — a living document of each model's known strengths, weaknesses, cost profile, and recent performance — to determine the optimal routing decision. For complex tasks, the orchestrator constructs multi-model chains: one model for reasoning, another for code generation, a third for synthesis. The user sees only one voice interface and one coherent response. The model complexity is entirely abstracted away — the right AI for the job, automatically.

Key Technical Decisions

  • Elasticsearch as capability registry — enables semantic search over model capabilities, not just keyword lookup
  • Living registry design — updated continuously as models release new versions or shift in performance
  • Orchestrator model as router — routing logic encoded in an LLM, not brittle static rules
  • Multi-model chaining support — architecture designed for tasks that benefit from specialized model combinations
  • Single voice interface — all routing complexity hidden from the user by design

Tech Stack

TypeScriptElasticsearchMultiple LLM APIsVoice AISemantic Search

Links