智能代理操作系统 是 AI Skill Hub 本期精选Agent工作流之一。综合评分 7.5 分,整体质量较高。我们推荐使用将其纳入你的 AI 工具库,帮助提升工作效率。
智能代理操作系统 是一套完整的 AI Agent 自动化工作流方案。通过可视化的节点编排,将复杂的多步骤任务拆解为清晰的自动化流程,实现全程无人值守的智能处理。支持与数百种外部服务和 API 无缝集成,适合构建数据处理管线、业务自动化和 AI 辅助决策系统。
智能代理操作系统 是一套完整的 AI Agent 自动化工作流方案。通过可视化的节点编排,将复杂的多步骤任务拆解为清晰的自动化流程,实现全程无人值守的智能处理。支持与数百种外部服务和 API 无缝集成,适合构建数据处理管线、业务自动化和 AI 辅助决策系统。
# 方式一:pip 安装(推荐)
pip install tinyagentos
# 方式二:虚拟环境安装(推荐生产环境)
python -m venv .venv
source .venv/bin/activate # Windows: .venv\Scripts\activate
pip install tinyagentos
# 方式三:从源码安装(获取最新功能)
git clone https://github.com/jaylfc/tinyagentos
cd tinyagentos
pip install -e .
# 验证安装
python -c "import tinyagentos; print('安装成功')"
# 命令行使用
tinyagentos --help
# 基本用法
tinyagentos input_file -o output_file
# Python 代码中调用
import tinyagentos
# 示例
result = tinyagentos.process("input")
print(result)
# tinyagentos 配置文件示例(config.yml) app: name: "tinyagentos" debug: false log_level: "INFO" # 运行时指定配置文件 tinyagentos --config config.yml # 或通过环境变量配置 export TINYAGENTOS_API_KEY="your-key" export TINYAGENTOS_OUTPUT_DIR="./output"
<p align="center"> <img src="static/taos-logo.png" alt="taOS" width="400"> </p>
Features unlock automatically based on your hardware and cluster. Solo Pi sees core features. Add a GPU worker and image generation, video, and training appear. No configuration, the platform just knows what's possible.
curl -fsSL https://raw.githubusercontent.com/jaylfc/tinyagentos/master/scripts/install-server.sh | sudo bash
Run without `sudo` to install as a user-mode systemd unit instead. The script is idempotent, safe to re-run on an existing install. Supports env-var overrides for install path, branch, and port.
**Manual / development:**
bash pip install -e . python -m uvicorn tinyagentos.app:create_app --factory --host 0.0.0.0 --port 6969 ```
Open http://your-host:6969 (or http://taos.local:6969 with mDNS). The root URL loads the desktop shell directly.
curl -fsSL https://raw.githubusercontent.com/jaylfc/tinyagentos/master/scripts/install-worker.sh | sudo bash -s -- http://your-server:6969
curl -sL https://raw.githubusercontent.com/jaylfc/tinyagentos/master/tinyagentos/worker/android_setup.sh | bash
powershell
5-step wizard: pick framework → choose model → configure → deploy into an isolated container (LXC on bare metal, Docker on VPS, auto-detected). Each agent gets its own memory system (taOSmd instance), its own file storage, and its own network identity. The framework runs inside the container but TinyAgentOS manages everything around it: memory, channels, secrets, model access, scheduled tasks, and inter-agent communication. This means the framework is a swappable component, not a lock-in decision.
**Running taOS inside an LXC (e.g. Proxmox)?** Deploying an agent creates a nested container, which an unprivileged LXC cannot do — the kernel can't remap the nested container's filesystem, so the deploy fails with anidmapped storage / change ownershiperror. Run the taOS LXC as privileged with nesting enabled. On Proxmox: untick Unprivileged container and set Options → Features →nesting=1(pluskeyctl=1,fuse=1), then redeploy. Bare-metal and VM installs are unaffected. (taOS detects this and surfaces the fix in the deploy error.)
<p align="center"> <img src="docs/images/mobile-agents-empty.png" alt="Agents app empty state on mobile — one tap to deploy" width="30%"> </p>
<p align="center"><sub>The Agents app on mobile — one tap from empty to your first deployed agent.</sub></p>
Full transparency on every file, service, user, and port the installers touch. Nothing is hidden behind a vendored binary; everything is plain Python, plain systemd, plain shell.
Run curl -fsSL https://raw.githubusercontent.com/jaylfc/tinyagentos/master/scripts/install-server.sh | sudo bash on a fresh Debian / Ubuntu / Fedora / Arch / Alpine box to get the controller fully installed, repo cloned to ~/tinyagentos/, venv created, all deps installed, and both tinyagentos.service (port 6969) and qmd.service (port 7832) registered and started.
| Where | What |
|---|---|
/etc/systemd/system/tinyagentos.service | Main controller systemd unit. Runs uvicorn on port 6969. |
/etc/systemd/system/qmd.service | Embedding backend (embed / rerank / query expansion) on port 7832. Used by taOSmd for vector operations. Backed by rkllama on RK3588 boards or local node-llama-cpp elsewhere. |
tinyagentos-sdcpp.service (repo root) | (RK3588 only) CPU image generation backend. Manual setup only — not auto-installed by the installer. |
tinyagentos-rknn-sd.service (repo root) | (RK3588 only) NPU image generation backend. Manual setup only — not auto-installed by the installer. |
/home/<user>/tinyagentos/ | The repo checkout. All code, all configs. |
/home/<user>/tinyagentos/.venv/ | Python virtualenv. All Python deps live here, never pip install to system Python. |
/home/<user>/tinyagentos/data/ | All persistent state. **One directory to back up.** Contains: agent state YAMLs, agent memory SQLite indexes, agent workspaces, secrets DB, scheduler history, channel credentials, downloaded models, torrent settings, telemetry opt-in flag. |
/home/<user>/.cache/qmd/index.sqlite | User memory index (taOSmd knowledge base for personal notes). Per-agent indexes live separately under data/agent-memory/{name}/index.sqlite. |
| Ports listened on | **6969** (controller HTTP API + web UI), **6970** (browser-proxy second-origin, TAOS_BROWSER_PROXY_PORT), **7832** (qmd embedding service), **4000** (LiteLLM proxy, localhost only by default) |
| OS packages added | python3 + venv + pip, git, curl, ca-certificates, libtorrent-rasterbar (model torrent mesh), Node.js 22 (qmd + SPA build), sqlite3, libsqlcipher (encrypted secrets), vulkan-tools (hardware detection), postgresql (LiteLLM virtual keys) |
| User accounts created | The distro postgres system user is created when PostgreSQL is installed. A litellm Postgres role and database are created for virtual-key management. Everything else runs as the user who ran the installer. |
| Where | What |
|---|---|
/etc/systemd/system/tinyagentos-worker.service | Worker systemd unit (when run with sudo). Connects to the controller URL you passed and registers this machine as a cluster node. Runs as your user via User=, not root. |
~/.config/systemd/user/tinyagentos-worker.service | Same unit as above, but in user-mode (when run without sudo). Linger is enabled automatically. |
~/.local/share/tinyagentos-worker/ | Worker repo checkout. ~150 MB on disk after install. Self-contained venv inside. |
~/.local/share/tinyagentos-worker/.venv/ | Python venv with worker-only deps: httpx, pydantic, psutil, fastapi, uvicorn, pyyaml, pillow, libtorrent. **Does NOT install controller-side deps** (no aiosqlite, no LiteLLM, no scheduler engine). |
| Ports listened on | None. Workers are pure outbound, they connect TO the controller. |
| OS packages added | python3, venv, pip, git, curl, ca-certificates, libtorrent (Debian/Ubuntu only, Arch/Fedora/Alpine equivalents on those distros). |
| User accounts created | None. The worker runs as the user who ran the installer. |
```bash
systemctl status tinyagentos-worker ls ~/.local/share/tinyagentos-worker/ ```
```bash
Controller (recommended): use the one-shot update script:
cd ~/tinyagentos
bin/update.sh
This pulls the latest, rebuilds the desktop frontend bundle if the source has moved (skips it when nothing changed), then restarts the service. The frontend rebuild takes ~50s when it fires; it is a no-op otherwise.
Manual equivalent:
```bash cd ~/tinyagentos git pull
find . -name pycache -type d -exec rm -rf {} + 2>/dev/null || true
cd desktop && npm install && npm run build && cd .. sudo systemctl restart tinyagentos
The systemd unit also runs a conditional rebuild as an `ExecStartPre` step — if you skip the manual `npm run build`, the next service restart detects the stale bundle and rebuilds it automatically (~50s startup overhead when it fires).
**Worker:**
bash cd ~/.local/share/tinyagentos-worker git pull find . -name pycache -type d -exec rm -rf {} + 2>/dev/null || true sudo systemctl restart tinyagentos-worker ```
The bytecode cleanup line is belt-and-braces; Python's mtime-based invalidation usually works, but on long-running boxes that have survived many upgrades it occasionally doesn't, and a stale .pyc is easy to mistake for a code bug.
scripts/install-rknpu.sh is an opt-in automated installer for the full Rockchip NPU stack. It pins librknnrt to 2.3.0, installs the jaylfc fork of rkllama, and preloads three chat models. All binaries are fetched from huggingface.co/jaysom/tinyagentos-rockchip-mirror, a TAOS-controlled mirror, and SHA256-verified before installation. If any checksum fails the script hard-aborts.
curl -fsSL https://raw.githubusercontent.com/jaylfc/tinyagentos/master/scripts/install-rknpu.sh | sudo bash
See docs/mirror-policy.md for the mirror governance policy, what is mirrored, when it updates, how to verify integrity independently, and how to self-host the mirror for air-gapped deployments. The same policy will extend to RK3576, Raspberry Pi 4, Mac mini / Apple Silicon, and x86 classes as those verified install paths land.
Controller (server):
```bash
Framework-agnostic skill system with 8 default skills, memory_search, file_read, file_write, web_search, code_exec, image_generation, list_image_models, http_request, categorised by search, files, code, media, browser, data, comms, system. Each skill declares compatibility per framework (native/adapter/unsupported) and works across all 16 supported frameworks via adapter translation. Assign or remove skills per agent from the Skills tab with compatibility badges.
One-click install for agent frameworks, AI models, and services. Hardware-aware, only shows what works on your device.
app-catalog/plugins/ ships 47 MCP servers including the official set (filesystem, git, fetch, memory, sequential-thinking, time), GitHub, Playwright, Docker, Kubernetes, databases (Postgres/MySQL/SQLite dbhub, MongoDB, Redis, Chroma, Supabase), documents (pandoc, office docs, spreadsheet, markdownify, excel), comms (Slack, WhatsApp, email, Notion, Obsidian, Atlassian, Google Workspace), infra (AWS, Cloudflare, Grafana, arXiv, YouTube transcript, Firecrawl), agent-specific (browser-use, Camoufox, context7, supergateway, engram, Exa), Home Assistant, Todoist, and more.
| Category | Apps |
|---|---|
| **Agent Frameworks (15)** | SmolAgents, PocketFlow, OpenClaw, nanoclaw, PicoClaw, ZeroClaw, MicroClaw, IronClaw, NullClaw, Moltis, Hermes, Agent Zero, OpenAI Agents SDK, Langroid, ShibaClaw |
| **Streaming Apps (13)** | Blender, LibreOffice, Code Server, GIMP, Krita, FreeCAD, Obsidian, Excalidraw, JupyterLab, Grafana, n8n, Terminal, Neko Browser |
| **LLM Models** | 112-manifest local catalog: Qwen3 0.6B-32B, Qwen2.5 0.5B-72B (+ RKLLM 1.5B-14B for RK3588), Llama 3.1/3.2/3.3, Gemma 2/3, Phi-3.5/4/4-mini, Mistral/Nemo/Mixtral, DeepSeek, Granite, Command-R, SmolLM2, TinyLlama, plus 167k+ searchable from HuggingFace |
| **Vision Models** | Qwen2-VL, Qwen2.5-VL, MiniCPM-V 2.6, Moondream2, Florence-2, LLaVA 1.6 / LLaVA-Phi-3 |
| **Embeddings / Rerankers** | nomic-embed-text-v1.5, bge-large/small/m3, mxbai-embed-large, snowflake-arctic-embed, qwen3-embedding/reranker, bge-reranker-v2-m3 |
| **Audio Models** | Whisper tiny→large-v3-turbo, Kokoro TTS, Piper voices, Parakeet TDT |
| **Image Models** | SD 1.5 LCM, Dreamshaper 8 LCM, LCM Dreamshaper V7 (+ RKNN for RK3588), SDXL Turbo/Lightning, Flux schnell/dev GGUF, SD 3.5 Large Turbo, PixArt-Σ, SDXS, Playground v2.5, Kolors, AuraFlow, Stable Cascade |
| **Image Tools** | RMBG-1.4, BiRefNet, Real-ESRGAN x4, 4x-UltraSharp, GFPGAN, CodeFormer, ControlNet (canny/depth/openpose) |
| **Image Gen** | ComfyUI, Fooocus, SD Web UI, stable-diffusion.cpp, FastSD CPU, RKNN SD, rk-llama.cpp |
| **Video Gen** | WanGP (Wan 2.1/2.2, HunyuanVideo), LTX Video |
| **Voice/Audio** | Whisper STT, Piper TTS, Kokoro TTS, Chatterbox, MusicGPT |
| **AI Tools** | Perplexica (AI search), Open WebUI, Dify, SearXNG |
| **Infrastructure** | Gitea, Code Server, n8n, Docker Mailserver, Tailscale, Dynamic DNS |
| **Home & Monitoring** | Home Assistant, Uptime Kuma, File Browser, Excalidraw, Memos, Linkwarden |
高质量的开源AI工作流项目,具有较高的实用价值
该工具使用 AGPL-3.0 协议,商用场景请仔细阅读协议条款,必要时咨询法律意见。
AI Skill Hub 为第三方内容聚合平台,本页面信息基于公开数据整理,不对工具功能和质量作任何法律背书。
建议在沙箱或测试环境中充分验证后,再部署至生产环境,并做好必要的安全评估。
⚠️ AGPL 3.0 — 最严格的 Copyleft,网络服务端使用也需开源,SaaS 使用受限。
经综合评估,智能代理操作系统 在Agent工作流赛道中表现稳健,质量良好。如果你已有明确的使用需求,可以直接上手体验;如果还在评估阶段,建议对比同类工具后再做决策。
| 原始名称 | tinyagentos |
| 原始描述 | 开源AI工作流:Self-hosted auto clustering AI agent OS for consumer hardware like the computer 。⭐198 · Python |
| Topics | ai-agentsai-platformdistributed-computingpython |
| GitHub | https://github.com/jaylfc/tinyagentos |
| License | AGPL-3.0 |
| 语言 | Python |
收录时间:2026-06-03 · 更新时间:2026-06-03 · License:AGPL-3.0 · AI Skill Hub 不对第三方内容的准确性作法律背书。
选择 Agent 类型,复制安装指令后粘贴到对应客户端