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

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
Jina AI
Jina AI
Hugging Face - Blog
Hugging Face - Blog
博客园 - 三生石上(FineUI控件)
博客园 - 【当耐特】
大猫的无限游戏
大猫的无限游戏
IT之家
IT之家
宝玉的分享
宝玉的分享
WordPress大学
WordPress大学
有赞技术团队
有赞技术团队
Apple Machine Learning Research
Apple Machine Learning Research
酷 壳 – CoolShell
酷 壳 – CoolShell
阮一峰的网络日志
阮一峰的网络日志
奇客Solidot–传递最新科技情报
奇客Solidot–传递最新科技情报
爱范儿
爱范儿
小众软件
小众软件
OSCHINA 社区最新新闻
OSCHINA 社区最新新闻
The Cloudflare Blog
S
SegmentFault 最新的问题
博客园 - Franky
博客园_首页
T
Tailwind CSS Blog
雷峰网
雷峰网
罗磊的独立博客

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 - Loxia-ai/onbuzz-community: OnBuzz - All-in-one &...
Dantlv · 2026-06-26 · via Hacker News - Newest: "AI"

🎯 Where to go first

🌐 Official site onbuzz.loxia.ai — downloads, news, showcase
🎓 Training onbuzz.loxia.ai/training — learn how to drive a fleet of agents end-to-end
🗺 Roadmap & task board roadmap.onbuzz.loxia.ai — what's planned, in flight, and open for contributors
📦 Releases github.com/Loxia-ai/onbuzz-community/releases — installers, binaries, and Electron desktop apps

If you're trying OnBuzz for the first time, go to onbuzz.loxia.ai/training. The training walks you through your first agent, your first flow, and your first multi-agent collaboration in under 30 minutes.

If you want to contribute, head to roadmap.onbuzz.loxia.ai to find a task to claim, then read CONTRIBUTING.md for the workflow.


What it does

OnBuzz Community is an open-source platform for running and orchestrating multiple AI agents locally. Each agent has its own workspace, tools, and conversation history, and they can talk to each other to solve bigger problems together.

It's the same engine that powers Loxia's commercial Autopilot product, re-released under Apache-2.0 with all server dependencies removed. You bring your own provider keys; OnBuzz talks to OpenAI / Anthropic / Gemini / xAI / Ollama directly. Nothing flows through anyone else's servers.

It runs as a local app with two interfaces:

  • A web UI in your browser (the default) — chat, agent management, flows, scheduled tasks.
  • A terminal UI — the same thing in a TUI for keyboard-driven workflows.

Why it's different

Most "AI assistant" tools are either a single chat window with a prompt, or a closed cloud product where your data lives on someone else's infrastructure. OnBuzz Community is built on three opinionated choices:

1. Local-first and zero-cloud by design. There is no central backend. There is no telemetry. Your conversations, agents, attachments, and credentials live in ~/.local/share/onbuzz (or your platform's equivalent). Disconnect your machine from the internet and Ollama-backed agents keep working.

2. Multi-agent, not single-chat. Agents are first-class. Each has a name, a system prompt, a working directory, and a configurable subset of tools. A "Coder" agent and a "Reviewer" agent can collaborate on the same repo in parallel, send each other messages, hand off tasks, and you can watch the whole thing run.

3. Bring your own keys, plug in any provider. Five providers ship out of the box — OpenAI, Anthropic, Gemini, xAI, and Ollama. You can also wire any OpenAI-compatible endpoint (OpenRouter, Together, Fireworks, Groq, vLLM, LiteLLM, Azure OpenAI, self-hosted vLLM…) with a base URL and a key. The model catalog is auto-discovered from each vendor's /models endpoint, so new releases show up automatically.

It also ships with 20+ built-in agent tools for real software work:

Code & files terminal, filesystem, file tree, code search (seek), file content replace
Code analysis import analyzer, dependency resolver, static analysis (ESLint/Prettier/TS/Stylelint), clone detection, code map
Web Puppeteer-driven browser, web fetch, scrape, screenshot
Documents PDF, DOCX, XLSX read/write
Coordination task manager, agent-to-agent messaging, agent delays, job-done signaling
Authoring persistent memory, skills (reusable agent recipes), user prompts, visual editor
Integrations Discord / Telegram / WhatsApp bridges (BYO bot tokens, all optional)
Scheduling cron-like recurring tasks and flow runs

🚀 Quick start

💡 Prefer a guided walkthrough? The training program at onbuzz.loxia.ai/training runs you through these steps with screen recordings, troubleshooting tips, and recipe templates. Recommended for first-time users.

1. Install

npm install -g onbuzz-community

Or grab a pre-built installer / binary from github.com/Loxia-ai/onbuzz-community/releases (Windows, macOS, Linux — CLI, native installer, or Electron desktop app).

(Requires Node.js ≥ 20 if installing via npm.)

2. Add a provider key

Either:

  • Run onbuzz web, open Settings, paste a key from any of: OpenAI / Anthropic / Gemini / xAI; or
  • Install Ollama and ollama pull llama3.1:8b for fully-offline use (no key needed).

3. Start chatting

Your browser opens, you create your first agent, send a message — done. From there you can spin up more agents, give them their own working directories, and let them collaborate.

Prefer the terminal? onbuzz plus-terminal starts the server and a TUI together.


💡 Example use cases

People use OnBuzz Community for things like:

  • Coding tasks at scale — give an agent a repo and a goal ("audit my dependencies for vulnerabilities and open PRs") and let it work autonomously while you do something else.
  • Research workflows — multi-agent flows where one agent gathers sources, another summarizes, another fact-checks.
  • Document pipelines — extract structured data from a folder of PDFs, generate Excel reports, write a DOCX summary.
  • Personal knowledge work — long-running agents with persistent memory that learn your preferences and codebase over weeks.
  • Bot orchestration — wire agents to Discord/Telegram/WhatsApp so your team or community can address them naturally.
  • Local privacy-first AI — using Ollama, run everything on your laptop. Sensitive client work never leaves the machine.
  • Provider experimentation — run the same prompt across GPT-5, Claude Opus, Gemini, Grok, and a local Llama in parallel. Compare cost, latency, quality.

🧱 Repository structure

onbuzz-community/
├── bin/                    CLI entry point (the `onbuzz` command)
├── config/                 Default model + benchmark manifests (override-friendly)
├── docs/                   Provider setup, architecture, contribution guides
├── electron/               Optional Electron desktop app shell
├── installer/              NSIS / PKG / DEB installer scripts
├── scripts/                Build + maintenance helpers
├── skills/                 Bundled example skills (e.g. web-game-dev)
├── src/
│   ├── core/               Orchestrator, agent pool, scheduler, flow executor
│   ├── interfaces/         HTTP/WebSocket server + terminal UI client
│   ├── services/
│   │   ├── providers/      LLM provider adapters (one per vendor)
│   │   ├── aiService.js    Dispatcher that routes to providers
│   │   └── …               Models, benchmarks, scheduling, memory, etc.
│   ├── tools/              Agent-facing tools (terminal, filesystem, web, …)
│   └── utilities/          Shared helpers
├── web-ui/                 React frontend (Vite-built)
└── .github/                CI workflows + issue templates

The high-level flow is: Web/Terminal UI → WebSocket → OrchestratorAgent Pool & SchedulerMessage ProcessorAIServiceProvider Registry → vendor API.


🔧 Configuration

Most things just work. A few env vars are useful:

Variable Purpose
OPENAI_API_KEY / ANTHROPIC_API_KEY / GEMINI_API_KEY / XAI_API_KEY Provider keys (or paste in Settings UI)
OLLAMA_HOST Ollama daemon URL (default http://127.0.0.1:11434)
LOXIA_PORT / LOXIA_HOST Web server port and host
LOXIA_LOG_LEVEL debug / info / warn / error
LOXIA_MODELS_PATH Override the default model manifest with a custom JSON file
LOXIA_BENCHMARKS_PATH Override the routing-benchmark text

User overrides also work without env vars: drop ~/.onbuzz/models.json or ~/.onbuzz/benchmarks.json and OnBuzz will pick them up.

For per-provider details (auth headers, special models, OpenAI-compatible custom endpoints), see docs/PROVIDERS.md. For the built-in task, message, delay, and completion flow agents use to coordinate, see docs/AGENT_COORDINATION.md.


🤝 Contributing

OnBuzz Community is open to contributions of every size — bug fixes, new tools, new provider adapters, docs, examples, or "I tried this and the install was confusing" feedback.

The contribution path:

  1. Find a task on the public roadmap & task board, or browse the 12-week dev program.
  2. Claim it by commenting on the linked issue.
  3. Read CONTRIBUTING.md for the full workflow — fork → branch → PR → review → squash-merge, with details on branch naming, Conventional Commits, rebasing, and code-review etiquette.

Looking for a first thing to work on? Our 12-week community program is a menu of 48 fully-specified tasks — each with context, a definition-of-done, deliverables, and pointers into the codebase. Daily tasks are sized for a single sitting; weekly capstones are larger pieces of work. Audience tracks span developers, designers, writers, QA, security, DevOps, data scientists, localizers, DevRel, accessibility advocates, visual designers, and domain experts.

Not a developer? Your domain knowledge is exactly what we need. docs/RECIPES_BY_FIELD.md has a deep menu of contribution ideas for teachers, clinicians, accountants, content creators, novelists, real estate agents, lawyers, restaurant owners, GMs, and many more — none of which require writing code.

If you find a security issue, please don't file a public issue — email contact@loxia.ai instead.


🔒 Privacy

  • Local-first — your conversation history, agent state, attachments, and all generated artifacts stay on your machine.
  • No telemetry — OnBuzz Community does not phone home. The only outbound network calls are to the LLM provider you configure.
  • Encrypted credentials — API keys are stored AES-256-GCM-encrypted under your user-data directory, derived from a machine-specific identifier.
  • Easy reset — delete the user-data directory to wipe all local state.

📄 License

Apache License 2.0 — see also NOTICE.

Copyright © 2025–2026 Loxia Labs LLC and OnBuzz Community contributors.

OnBuzz Community is forked from the proprietary Loxia Autopilot codebase. The commercial Autopilot product continues separately, with hosted models, a marketplace, and managed updates. If you want those things, see autopilot.loxia.ai. If you want to run agents yourself, with your keys, on your hardware, with full source access — you're in the right place.