Phi Browser

# AI Engineer## CategoryEngineering## Available in* San Francisco* London* Beijing* Shanghai* Shenzhen* Qingdao## About PhinomenonPhinomenon is built by browser and cloud veterans: our founders are behind FydeOS (a Chromium-based OS trusted by millions) and QingCloud (one of China's earliest API-first IaaS platforms, founded in 2012). Fresh off a healthy, eight-figure Series A raise, we have the runway to dream big without panicking about next week's payroll. With folks working from San Francisco, London, Beijing, Shanghai and Shenzhen, we are proudly global and friendly to your time zone, with overlapping hours so we can all actually talk. We ship fast and ship hard, but burnout is so last decade: we bake in wellness stipends, mandatory recharge days and flexible schedules, because great code comes from healthy lives. No timecards, no surveillance software, no weekend team-building. We'd rather hand you the best tools money can buy (Claude Max, ChatGPT Business and Cursor Ultra come standard) and trust you with them. We don't fetishise years of experience, degrees or big-tech logos, and fresh graduates are more than welcome: show us what you have built and how you think. If you want to build the nextgen browser that actually, you know, gets you, while still enjoying your evenings and maintaining upright posture, welcome home.## About the RoleAs an AI Engineer at Phinomenon, you build the AI system behind Phi Browser, all of it. Agents are infrastructure too: the same person who designs phi-agent's planning loops, tool use, and record-to-skill workflows should understand the local-first memory and knowledge systems those agents draw on, and be able to ship the APIs, data layers, and UIs that carry them to users. You'll work with frontier models and open protocols (MCP and friends), engineer for on-device constraints where every megabyte is someone's laptop, and ship weekly into a product people live in all day. The goal is simple to say and hard to build: a browser that's genuinely on your side. Your browser friend, not another company's data funnel.## Responsibilities* Build phi-agent's core loops: planning, tool use, skill execution, and graceful recovery when the web fights back* Develop browser-native agent capabilities: tab and workspace control, page understanding, record-to-skill workflows* Design and build the memory and knowledge systems underneath: local-first storage, ingestion to clean Markdown, hybrid retrieval, and proactive resurfacing* Ship the full stack that carries it (APIs, data layers, and web UIs), owning features end to end* Integrate and evaluate frontier LLMs; own context engineering and the cost/latency/quality trade-offs, including BYO-LLM paths* Expose Phi's capabilities to third-party agents via MCP and emerging open standards* Harden the AI surface against prompt injection, adversarial pages, and content poisoning; safety is a feature, not a patch* Instrument everything: evals, telemetry, and benchmarks that prove the system is actually getting smarter## Requirements* Show us AI systems you've actually built: agents, RAG/retrieval pipelines, or LLM-powered products in real use. We don't count years, degrees, or logos, and a new grad with a serious agent project is very much in the running.* Strong TypeScript or Python, and the range to move between agent logic, storage and retrieval internals, and UI without waiting for someone else* Hands-on fluency with modern agent patterns: function calling, planning loops, memory, evaluation harnesses* Pragmatic model sense: you know when to reach for a frontier model, a small one, or no model at all* Respect for user data as a design constraint: exportability, reversibility, local-first performance on consumer hardware* A security-aware mindset: every webpage is potentially hostile input to your agent* A bias for shipping: weekly releases don't scare you; unvalidated cleverness does## Nice to Have* Experience with Chromium/CDP, Playwright, or other browser-automation internals* Familiarity with SQLite/FTS, embedded vector stores, or local-first architectures (CRDTs, file-sync)* Contributions to open-source agent frameworks or the MCP ecosystem* Experience building eval pipelines for non-deterministic systems* You've built and abandoned your own second-brain system, and know exactly whyAnd should the future prove entirely too exhausting, there is a switch to disable all AI features. No offense will be taken. #J-18808-Ljbffr