Perspective Intelligence mixes Hermes, Apple
- Techopolis’s Perspective Intelligence appears to be wiring Apple’s on-device Foundation Models into a broader agent stack, using Hermes-style orchestration to hand work around. - The key detail is architectural: Apple’s framework already supports tool calling, while Perspective’s own server and CLI expose Apple models through OpenAI-compatible endpoints. - That makes the app interesting beyond one demo — it hints at iOS agents that keep private work local but still reach outside.
A new kind of iPhone AI app is starting to come into focus. Not just “chat with one model,” and not just “pick local or cloud,” but a router that decides which system should handle which part of the job. Perspective Intelligence looks like one of the clearest examples yet. That matters because Apple finally gave developers direct access to its on-device language model, but Apple’s model is not the whole story. Some tasks want privacy and speed. Some want bigger external models, long-running agents, or broader tool use. The interesting move is not choosing one side. It’s stitching them together. Apple’s Foundation Models framework supports on-device generation and tool calling in iOS 26 and related platforms. Hermes Agent, from Nous Research, is built around a persistent, tool-using agent loop with memory and skills. ### What is Perspective actually building? Perspective Intelligence is an App Store app from Techopolis Online Solutions. The public app listing pitches it as a private, on-device assistant for chat, OCR, image description, writing, and voice features on Apple devices that support Apple Intelligence. The app says requests are processed locally with Apple’s Foundation Models framework and related on-device technologies. (developer.apple.com) But Techopolis has also published adjacent infrastructure that shows how the team thinks. Its open-source Perspective server exposes Apple’s on-device model as OpenAI- and Ollama-compatible local endpoints. Its CLI does something similar for Foundation Models and MLX models on a Mac. Basically, the company is not treating Apple’s model as a sealed box inside one app. It is treating it like a component in a larger system. ### Why does Hermes matter here? (apps.apple.com) Hermes is not just another model name. It is an agent framework. The pitch is long-running autonomy — memory across sessions, reusable skills, lots of built-in tools, and the ability to live on your own infrastructure instead of inside one chat window. That is a very different shape from Apple’s local model, which is optimized for on-device language tasks and tool use inside an app experience. (github.com) So if Perspective is mixing the two, the split is pretty intuitive. Apple handles the fast, private, local stuff. Hermes handles the agent layer — planning, memory, external tools, and possibly handoffs to non-Apple models. ### Why is that hybrid pattern a big deal? Because it solves the obvious tradeoff. Pure on-device AI is private and cheap to run, but weaker and more constrained. Pure cloud AI is more capable, but slower, pricier, and harder to square with Apple’s privacy-first pitch. A hybrid stack lets a developer keep routine work on the phone and escalate only when needed. (hermes-agent.nousresearch.com) Apple’s own framework almost invites this. The Foundation Models docs explicitly describe tool calling to search local or online data sources or call services inside the app. That means the local model can act as a front door, not just a final answer engine. ### So is this an App Store loophole? Not exactly. More like a test case. Apple already allows apps to call services and use tools. Perspective’s public materials also already frame the product as local-first, which fits Apple’s positioning. The gray area is how far an app can go in making an external agent feel native while still presenting itself as an Apple Intelligence-style local experience. I couldn’t verify the exact X post or implementation details behind the claimed Hermes integration, but the company’s published app listing and open-source server show the underlying pattern clearly: Apple local models on one side, standard AI endpoints and external orchestration on the other. (developer.apple.com) ### What’s the bottom line? The important thing here is not one indie app. It is the architecture. Perspective suggests what the next wave of Apple AI apps may look like — local model in the interface, agent framework behind the curtain, and selective cloud reach only when the task really needs it. (developer.apple.com) (apps.apple.com)