Respan, Ekai, Nasiko and b.ai reframe gateways as centralized control planes
- Respan, Ekai and Nasiko are all pitching a new role for AI gateways: not just proxying model calls, but centralizing routing, credentials, spend controls, tracing and agent operations. - Nasiko says its platform registers, routes, executes and monitors agents, while Ekai says its gateway tracks usage and costs across providers and supports delegated, revocable key access. - The pitch mirrors a broader shift in the market, with vendors like Cloudflare also recasting AI gateways as the single plane for keys, billing and routing. (blog.cloudflare.com)
AI gateways are being recast as the operating layer for agents, not just a relay for model requests. Respan, Ekai and Nasiko now describe one place to handle keys, routing, budgets and telemetry. (github.com) (docs.ekailabs.xyz) (nasiko.com) That is a shift from the older pattern where each service held its own provider key and retry logic. The new pitch is to move those controls into a shared gateway that sits between apps, agents and model vendors. (docs.ekailabs.xyz) (nasiko.com) Nasiko is the clearest example of that language. Its site says it helps teams “register, secure, route, and observe” agents, and its open-source project describes centralized management, intelligent routing and observability for AI agents. (nasiko.com) (github.com) Its router service adds the mechanics behind that claim. Nasiko’s GitHub docs describe vector-based agent selection, fallback mechanisms, health checks, caching and a router orchestrator that coordinates the workflow. (github.com) Ekai is making a similar argument from the model and context side. Its docs call the product a “unified gateway and memory system,” with one endpoint for OpenAI, Anthropic, Google, xAI and OpenRouter. (docs.ekailabs.xyz) Ekai says the gateway tracks requests, tokens and costs per provider, and its Oasis writeup says the control plane stores encrypted API keys on Oasis Sapphire with delegated access, model restrictions, spending limits and instant revocation. (docs.ekailabs.xyz) (oasis.net) Respan’s public materials are thinner, but its repository shows the same consolidation instinct. Its README says a single Respan API key authenticates both LLM inference through its proxy and telemetry export for tracing. (github.com) That means the gateway is no longer framed as only a traffic cop. It becomes the place where teams can enforce who may call which model, how usage is tracked, and what happens when a provider fails or costs spike. (github.com) (docs.ekailabs.xyz) (github.com) The broader market is moving the same way. In August 2025, Cloudflare said its AI Gateway was becoming the place to control costs and API keys, route between providers and manage AI traffic, with access to more than 350 models across six providers. (blog.cloudflare.com) The common thread is centralization. As agent stacks spread across multiple models, tools and runtimes, vendors are betting the gateway becomes the control plane that keeps the whole system governed from one endpoint. (nasiko.com) (docs.ekailabs.xyz)
