{"slug": "introducing-zixcai-an-agentic-ai-chat-interface-powered-by-a-770b-parameter-80b", "title": "Introducing ZixcAI: An Agentic AI Chat Interface Powered by a 770B Parameter Model (80B Active)", "summary": "A developer launched ZixcAI, an agentic AI chat interface built on a custom 770-billion-parameter mixture-of-experts model with 80 billion active parameters per token. The system pairs a native PHP and SQLite backend with a multi-round tool-calling agent loop featuring retries, circuit breakers, per-turn audit logs, sandboxed code execution, and a separate vision sub-agent. The developer argues the 80B-active MoE design delivers roughly the per-token compute of a dense 80B model while carrying about 10x the total knowledge.", "body_md": "Today I'm launching [ZixcAI](https://zixcai.n4tzzofficial.my.id) — an agentic AI chat interface built on top of a custom **770 billion parameter mixture-of-experts model** with **80 billion active parameters per token**.\n\nMost \"AI chat\" products you see in 2026 are wrappers around an existing API. ZixcAI is not. The interface is mine, the agent runtime is mine, and the model routing is mine. This article is a walkthrough of what it is, how the model architecture works, and why the 80B-active design is the right call for an agentic product.\n\nZixcAI is an **agentic AI assistant** — meaning it doesn't just chat, it acts.\n\nWhen you send a message, the model can decide to:\n\nThe interface is a chat app — dark, minimal, keyboard-first. But the backend is an **agent loop**: a multi-round, tool-calling execution engine with retries, circuit breakers, and per-turn audit logs.\n\nIf that sounds like a lot, it's because it is. ZixcAI is not a small project.\n\nThe core of ZixcAI is a **mixture-of-experts (MoE) transformer** with:\n\nA dense 770B model would be brutal:\n\n| Approach | Total Params | Active/Token | Compute per Token | Memory Footprint | \n|---|---|---|---|---|\n| Dense 770B | 770B | 770B | Very high | Very high | \n| MoE 770B/80B | 770B | 80B | ~10% of dense | High but manageable | \n\nMoE gives you the **capacity of a giant model** with the **compute cost of a much smaller one**. You pay for memory (all experts must be loaded), but you don't pay for compute on every token.\n\nThis is the same principle behind models like Mixtral, DeepSeek-MoE, and GPT-4-class systems. It's the dominant architecture for frontier models in 2026 for a reason.\n\n80B active parameters is roughly the same compute footprint as a dense 80B model per token. That's a **serious** amount of reasoning capacity — comparable to or exceeding most \"70B\" models that people run for serious work.\n\nBut because it's MoE, ZixcAI has **10x the total knowledge** of a dense 80B model. Different experts specialize in different things — code, math, prose, tool use — and the router picks the right ones per token.\n\nFor an agentic product where the model needs to reason about tools, plan multi-step actions, and write code, this combination is ideal.\n\nEvery response streams token-by-token via SSE. Reasoning is displayed in a separate \"thinking\" panel you can expand. Tool calls appear inline as animated cards with status indicators.\n\nWhen ZixcAI runs code, it actually runs code — inside a sandboxed environment on the backend. You can see the stdout, stderr, exit code, and execution time. Files it creates stay in your workspace.\n\nSend an image, ZixcAI routes it to a vision sub-agent. The main reasoning model stays text-only (for speed and cost), but you get image understanding where it matters. This is a **deliberate architecture choice** — see below for why.\n\nZixcAI remembers facts about you across conversations. Tell it your name, your stack, your preferences — next session, it knows. Memory is user-visible and user-editable.\n\nEvery message has a \"fork\" button. Fork creates a new conversation starting from that message. You can explore tangents without polluting your main thread.\n\nCode blocks get full syntax highlighting, one-click copy, and download. Math renders via KaTeX — inline and block, matrices, aligned equations, piecewise functions, all supported.\n\nA ZixcAI turn looks like this:\n\nUser message\n\n↓\n\n[PHP API] → creates turn record → returns SSE meta event\n\n↓\n\n[Agent loop]\n\n├─ Build history (system prompt + memory + recent turns)\n\n├─ Call MoE model with tool schemas\n\n├─ If tool call: execute → append result → loop\n\n├─ If no tool call: stream final text to client\n\n↓\n\n[Persistence] → save turn, tool calls, usage, memory updates\n\ntext\n\nEvery layer is instrumented. Tool calls have circuit breakers. Streaming has watchdog timers. Every turn has a full audit trail.\n\nThe backend is **native PHP** with SQLite. I wrote about that choice here — it's an unorthodox stack for AI, but it works.\n\nOne thing that surprises people: **ZixcAI's main model is text-only**.\n\nBut ZixcAI understands images. How?\n\nWhen you attach an image, ZixcAI:\n\nWhy this architecture instead of a single multimodal model?\n\n**Cost.** Multimodal inference is expensive. If 95% of your traffic is text, paying multimodal cost on every request is wasteful.\n\n**Latency.** The main model stays fast. Vision runs in parallel when needed, then gets injected as text.\n\n**Flexibility.** You can swap the vision model without touching the main model. You can route specific image types to specific vision models. The main model never needs retraining when vision upgrades.\n\nThis is a pattern I expect to see more of as agentic products get more serious about unit economics.\n\nZixcAI is **not**:\n\nBut it's a real product built by one developer, running real tools, on a real MoE backend.\n\n[ZixcAI](https://zixcai.n4tzzofficial.my.id) is live. Sign up, send a message, ask it to write code, upload an image, ask it to remember something. It's free to try.\n\nIf you're building agentic products, I'm curious what your stack looks like:\n\nDrop a comment. I read every one.\n\n*ZixcAI is built by [@N4tzzOfficial](https://github.com/n4tzz). Follow for updates as I ship new features — memory improvements, faster streaming, and more tool primitives.*", "url": "https://wpnews.pro/news/introducing-zixcai-an-agentic-ai-chat-interface-powered-by-a-770b-parameter-80b", "canonical_source": "https://dev.to/n4tzzofficial/introducing-zixcai-an-agentic-ai-chat-interface-powered-by-a-770b-parameter-model-80b-active-33oi", "published_at": "2026-09-26 03:53:21+00:00", "updated_at": "2026-09-26 04:30:02.318654+00:00", "lang": "en", "topics": ["artificial-intelligence", "large-language-models", "ai-agents", "ai-products", "ai-tools"], "entities": ["ZixcAI", "PHP", "SQLite", "Mixtral", "DeepSeek-MoE", "GPT-4"], "also_reported_by": [], "alternates": {"html": "https://wpnews.pro/news/introducing-zixcai-an-agentic-ai-chat-interface-powered-by-a-770b-parameter-80b", "markdown": "https://wpnews.pro/news/introducing-zixcai-an-agentic-ai-chat-interface-powered-by-a-770b-parameter-80b.md", "text": "https://wpnews.pro/news/introducing-zixcai-an-agentic-ai-chat-interface-powered-by-a-770b-parameter-80b.txt", "jsonld": "https://wpnews.pro/news/introducing-zixcai-an-agentic-ai-chat-interface-powered-by-a-770b-parameter-80b.jsonld"}}