惯性聚合 高效追踪和阅读你感兴趣的博客、新闻、科技资讯
阅读原文 在惯性聚合中打开

推荐订阅源

人人都是产品经理
人人都是产品经理
Stack Overflow Blog
Stack Overflow Blog
S
SegmentFault 最新的问题
博客园 - 司徒正美
aimingoo的专栏
aimingoo的专栏
U
Unit 42
GbyAI
GbyAI
B
Blog RSS Feed
博客园 - Franky
L
LangChain Blog
Hugging Face - Blog
Hugging Face - Blog
美团技术团队
The GitHub Blog
The GitHub Blog
Y
Y Combinator Blog
云风的 BLOG
云风的 BLOG
H
Hackread – Cybersecurity News, Data Breaches, AI and More
博客园 - 三生石上(FineUI控件)
Microsoft Azure Blog
Microsoft Azure Blog
钛媒体:引领未来商业与生活新知
钛媒体:引领未来商业与生活新知
G
Google Developers Blog
Last Week in AI
Last Week in AI
阮一峰的网络日志
阮一峰的网络日志
让小产品的独立变现更简单 - ezindie.com
让小产品的独立变现更简单 - ezindie.com
Apple Machine Learning Research
Apple Machine Learning Research

Hacker News - Newest: "AI"

AI can't read an investor deck AI as an attorney? Student uses ChatGPT, Gemini to sue UW over alleged racial discrimination Hacking MCP Servers in AI Systems – The Rug Pull: Tool Changes After Approval GitHub - MeepCastana/KubeezCut: Free Web based video editor Can AI judge journalism? A Thiel-backed startup says yes, even if it risks chilling whistleblowers Coming soon: 10 Things That Matter in AI Right Now DARPA built an AI to fact-check enemy weapons claims What explains heterogeneity in AI adoption? When AI Meets Muscle: Context-Aware Electrical Stimulation Promises a New Way to Guide Human Movements - Department of Computer Science AI Changed How We Build. It Did Not Change What Matters. Linux rules on using AI-generated code - Copilot is OK, but humans must take 'full responsibility for the… Meta spins up AI version of Mark Zuckerberg to engage with employees Code Mode: Let Your AI Write Programs, Not Just Call Tools | TanStack Blog GitHub - Delavalom/graft: Go framework for building AI agents. Type-safe tools, multi-provider (OpenAI, Anthropic, Gemini, Bedrock), zero vendor SDKs. India's TCS tops estimates, says new AI models did not dent services demand Gen Z's fading AI hype Strong feeling: we are in a folded AI reality GitHub - machinarii/total-recall-catalog: A reference catalog of latest knowledge retrieval, memory & RAG systems GitHub - mensfeld/code-on-incus: Give each AI agent its own isolated machine with root, Docker, and systemd. Active defense detects and stops threats automatically.. Quantization, LoRA, and the 8% Problem: Benchmarking Local LLMs for Production AI Iran war: We spoke to the man making Lego-style AI videos that experts say are powerful propaganda Powell, Bessent discussed Anthropic's Mythos AI cyber threat with major U.S. banks GitHub - immartian/bellamem: Persistent belief-graph memory for AI agents. Retrieves decisive context by importance — not recency, not RAG, not /compact. recursive-mode: The Repo-Native Operating System for AI Engineering After the attack on Sam Altman's home, will AI CEO's go on the offensive? The biggest advance in AI since the LLM Opus 4.6 vs GPT 5.4 One Prompt Unity World Generation Test “AI polls” are fake polls Client Challenge Can AI be a 'child of God'? Inside Anthropic's meeting with Christian leaders
GitHub - shreyasks094/Zeus: Self-hosted AI agent orchestr...
blackhawk094 · 2026-06-25 · via Hacker News - Newest: "AI"

⚡ Zeus — your one-stop tool to control your computer

Zeus is a local AI agent orchestrator with a real UI. Install once, run zeus start, and you get a fast, terminal-inspired web app — on your laptop and your phone — where AI agents can actually do things on your machine: read and write files, run commands, drive a real browser, search the web, manage tasks, run scheduled jobs, and orchestrate teams of sub-agents. One tool for everything you need to control your computer.

Everything lives in ~/.zeus. No cloud account, no telemetry, no lock-in — your agents, skills, memory, and workspaces are plain human-editable files on your own disk.

./install.sh     # one-time setup
zeus start       # launch the server + open the web UI

Screenshot 2026-06-24 at 20 29 52

Why Zeus is different

Most agent projects are a library or a terminal loop. Zeus is a complete application:

  • It has a real, beautiful UI — a vanilla-JS, no-build web app with a command palette, theming, live streaming, and inline HTML/SVG previews. Not a CLI you squint at.
  • It runs on your phone. The UI is an installable PWA with a mobile-first nav, and zeus start --tailscale puts it on your private tailnet so you can drive your computer's agents from anywhere — securely, never exposed to the public internet.
  • Agents control the whole machine, with a UI for it. Files, shell, a dedicated automated Chromium the agent drives by vision, web search, an embedded terminal, a file explorer, a code editor, git — all in one place.
  • It's an orchestrator, not just an agent. Sub-agent delegation (with per-sub-agent stop), recurring workflows, scheduled jobs, MCP servers, to-dos, dashboards, widgets, timers, and a layered, self-curating memory system.
  • Bring any model. Anthropic, OpenAI, Groq, Google Gemini, OpenRouter, AWS Bedrock — or fully local models via the built-in BYO Model app (a self-contained Ollama that installs entirely inside ~/.zeus).
  • Self-contained integrations. Need private web search or a managed browser? Zeus installs SearXNG, Chromium, and Ollama inside ~/.zeus — nothing touches your system.

Features

The zeus command

zeus start · stop · status · restart · logs -f · open · reset. Runs a detached local server (uvicorn on 127.0.0.1:8756) and opens the UI. zeus reset wipes ~/.zeus and starts factory-fresh (asks first).

Works on web and mobile

  • One-URL web UI at http://127.0.0.1:8756, served no-store so edits show up on reload.
  • Installable PWA with an offline page, app-launcher nav, and a touch-friendly mobile layout (centered popups, big tap targets, no horizontal scroll).
  • Tailscale built in: zeus start --tailscale serves the app on your tailnet (https://<machine>.ts.net) so your phone — logged into the same tailnet — can install and use Zeus from anywhere. Private by default.

Bring your own model

  • Cloud: Anthropic (API key), OpenAI, Groq, Google Gemini, OpenRouter, AWS Bedrock — paste a key and go. One model switcher, with per-model type badges (multimodal / text / tools) and editable context windows.
  • Local — the BYO Model app: browse the full Ollama library, favorite the models you like, pull any model (with live progress + a stop button), and run them fully on-device. The Ollama binary and weights live entirely under ~/.zeus/ollama on a private port — your system stays untouched. Vision models (e.g. qwen3-vl, llava) can see image attachments.

Agents, sub-agents & skills

  • Build specialized agents with their own prompt, tools, model, and permission mode.
  • Agents delegate to sub-agents (configurable depth) and you can stop a single sub-agent from the live dock while the parent keeps working.
  • A skills library the agent loads on demand; agents can even author new skills and sub-agents.

A computer the agent can actually use

  • Built-in tools: Read / Write / Edit / Bash / Glob / Grep, web search + fetch, and a vision-driven browser (a dedicated Chromium the agent clicks/types/scrolls by looking at screenshots).
  • A real embedded terminal (PTY + xterm.js), file explorer, code editor, and git panel.
  • Permission modesdefault (approve risky actions), plan (read-only), acceptEdits, autopilot — plus lifecycle hooks for PreToolUse / PostToolUse / Stop / UserPromptSubmit.

Automate everything

  • Workflows: multi-step agent pipelines on a cron schedule, on demand, or after every chat.
  • To-dos & stickies, timers & alarms, world clocks, live widgets & dashboards, desktop push notifications, and a built-in location tool.
  • MCP servers (stdio + HTTP/SSE) merge their tools into your agents.

Memory that stays sharp

  • Layered context — Soul / Steering / User / Project — loaded into every turn and prompt-cached so it's nearly free.
  • Model-driven curation: a model periodically de-duplicates and prunes memory itself, keeping a small, high-signal core. Project memory follows your working directory; volatile state (which tools/models are connected) is never stored.
  • Smart context management: dynamic per-model context windows, prompt caching, accurate token accounting, and in-turn compaction that clears stale tool outputs so long agentic runs don't blow past the window.

Self-contained integrations

SearXNG (private metasearch), a managed Chromium, and Ollama — each installs and runs entirely inside ~/.zeus, controllable from Settings → Integrations. Nothing is installed system-wide.

And more

A live performance monitor, themeable UI, a ⌘K command palette, per-tab working directories, multiple workspaces, image/PDF attachments, and spec-driven development tooling.


Quick start

./install.sh                 # creates backend/.venv, installs deps, symlinks `zeus` onto PATH
zeus start                   # start + open the UI at http://127.0.0.1:8756
zeus start --tailscale       # also serve on your tailnet (use it from your phone)

Then open the UI, connect a provider (or install a local model in BYO Model), and start chatting. Backend changes need zeus restart; the frontend is no-build — just reload.

Requirements: Python 3.11+ and macOS/Linux. State lives in ~/.zeus (override with ZEUS_HOME).


Acknowledgements

Huge thanks and inspiration to PewDiePie's Odysseus, openclaw, and hermes — projects that showed what personal, local-first AI agents can be. Zeus took the spark and ran in its own direction: a full app you can hold in your hand.

And enormous thanks to the open tools Zeus stands on:

  • Ollama — local models, installed and run entirely inside ~/.zeus.
  • SearXNG — private metasearch, self-hosted in ~/.zeus.
  • Chromium — the dedicated browser the agent drives by vision.
  • FastAPI & xterm.js — the backend and the embedded terminal.
  • And every model provider Zeus connects to — Anthropic, OpenAI, Groq, Google Gemini, OpenRouter, AWS Bedrock.

Zeus was vibe coded with passion — built for the joy of it. ⚡