OpenIntentOS

The Intent-Driven AI Operating System

Built entirely in Rust. <10MB binary. <200ms cold start. 7-provider cascade failover.
One binary. Say what you want. The OS handles the rest.

Documentation • Quick Start • Architecture

v0.1.10 — Early Release (March 2026)

OpenIntentOS is under active development. Core systems are production-stable. New adapters and capabilities ship weekly. Pin to a specific commit for stability until v1.0. Report issues here.

Install

macOS · Linux · WSL · Raspberry Pi · Android (Termux)

curl -fsSL https://raw.githubusercontent.com/OpenIntentOS/OpenIntentOS/main/install.sh | bash

Windows 10 / 11 — open PowerShell and run:

irm https://raw.githubusercontent.com/OpenIntentOS/OpenIntentOS/main/install.ps1 | iex

That's it. The installer will:

Auto-detect your OS and download the right prebuilt binary
Walk you through entering API keys — with a direct link for each one
Install a system service so the bot survives reboots and auto-restarts on crash
Start the bot immediately

No Rust, no Docker, no terminal experience required.

Platform	Architecture	Install command
macOS	Apple Silicon (M1/M2/M3/M4)	`curl ... \| bash`
macOS	Intel	`curl ... \| bash`
Linux	x86_64 (PC / VPS / WSL)	`curl ... \| bash`
Linux	ARM64 (Raspberry Pi 4/5 · AWS Graviton)	`curl ... \| bash`
Windows 10/11	x64	`irm ... \| iex`
Windows	ARM64 (Snapdragon laptops)	`irm ... \| iex`
Android	Termux (ARM64)	`curl ... \| bash`

No API Key? Start for Free

You don't need a paid subscription to get started. Pick any one:

🇨🇳 China — No VPN Required

Provider	Free Tier	Register
硅基流动 SiliconFlow ⭐	14M tokens/month, hosts DeepSeek/Qwen/GLM	siliconflow.cn
DeepSeek	¥10 free credit	platform.deepseek.com
Moonshot Kimi	Free trial credit	platform.moonshot.cn
智谱 GLM-4-Flash	Permanently free	open.bigmodel.cn

🌍 International

Provider	Free Tier	Register
NVIDIA NIM ⭐	1000 credits/month, 400B+ models	build.nvidia.com
Google Gemini	Generous free limits	aistudio.google.com/apikey
Groq	Free, fastest inference	console.groq.com

Run Locally (No Internet)

# Install Ollama
curl -fsSL https://ollama.com/install.sh | sh
ollama pull qwen2.5

# OpenIntentOS will auto-detect Ollama — no key needed
openintent serve

What is OpenIntentOS?

OpenIntentOS is an Intent-Driven AI Operating System — not a chatbot, not a Python wrapper around an LLM, not a "multi-agent framework."

Users express what they want in natural language. The system understands the intent, plans execution across tools, manages credentials securely, and delivers results — autonomously, with no babysitting required.

The entire system ships as a single ~10MB binary. No Node.js runtime, no Docker pull, no pip install.

What it looks like

User: Search GitHub for Rust async benchmarks, summarize the top 3 papers, save to /reports

→ web_search("Rust async runtime benchmarks 2025")           [0.8s]
→ web_fetch("https://arxiv.org/abs/...")                     [1.1s]
→ web_fetch("https://github.com/tokio-rs/tokio/discussions") [0.9s]
→ fs_write_file("reports/rust-async-benchmarks.md", ...)     [<1ms]
✓ Done — summary saved. 3 sources cited. 1,204 tokens used.

User: Check my unread emails, summarize anything urgent, post to the team Feishu group

→ email_list_inbox(unread=true, limit=20)                    [0.6s]
→ [analyzing 20 emails via ReAct reasoning]
→ feishu_send_message(group="Product Team", text="...")      [0.4s]
✓ Done — 3 urgent threads summarized, posted to Feishu.

User: 发一条抖音视频，标题是"周报回顾"，文件在 /tmp/weekly.mp4

→ douyin_publish_video(file_path="/tmp/weekly.mp4", caption="周报回顾")   [8.2s]
✓ 发布成功 — item_id: 7xxxxxxxxxxxxxxx

Cascading LLM Failover — Seven Providers, Zero Downtime

This is the core reliability feature. When your primary provider fails for any reason — rate limit, auth error, 503, model not found — OpenIntentOS automatically tries the next provider in the chain, transparently, in the same request.

Primary fails (429 / 401 / 404 / 502 / 503)
       │
       ├─► DeepSeek               deepseek-chat           (free credit)
       ├─► SiliconFlow 硅基流动   DeepSeek-V3             (free 14M/month, 🇨🇳)
       ├─► Moonshot Kimi           moonshot-v1-8k          (free credit, 🇨🇳)
       ├─► Zhipu GLM-4-Flash       glm-4-flash             (permanently free, 🇨🇳)
       ├─► Google Gemini           gemini-2.5-flash        (free tier)
       ├─► Groq                    llama-3.3-70b-versatile (free tier)
       ├─► NVIDIA NIM              qwen/qwen3.5-397b-a17b  (free tier)
       ├─► NVIDIA NIM              moonshotai/kimi-k2.5
       ├─► NVIDIA NIM              nvidia/nemotron-3-nano-30b-a3b
       └─► Ollama                  qwen2.5:latest          (local, offline)

Each provider has a 120-second cooldown after failure. After cascade exhaustion, the bot restores the primary and retries on the next message. Per-chat model overrides are preserved across failovers.

WARN  OpenAI 429 rate limit — starting cascade failover
INFO  trying NVIDIA NIM (qwen/qwen3.5-397b-a17b)
INFO  ✓ NVIDIA NIM responded in 1.2s — continuing on this provider
INFO  primary provider restored after 120s cooldown

OpenIntentOS vs The Landscape

All data from official documentation and public repositories — February 2026.

Cold Start Time (lower is better)

OpenIntentOS  ████░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░  <200ms    ★
LangGraph     █████████████████░░░░░░░░░░░░░░░░░░░░░   2.5s
CrewAI        ████████████████████░░░░░░░░░░░░░░░░░░░   3.0s
AutoGen       ██████████████████████████░░░░░░░░░░░░░   4.0s
OpenClaw      █████████████████████████████████████░░   ~6s

Binary / Install Size (lower is better)

OpenIntentOS  ██░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░   ~10MB    ★
CrewAI        ████████░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░  ~100MB
LangGraph     ████████████░░░░░░░░░░░░░░░░░░░░░░░░░░  ~150MB
AutoGen       ████████████████░░░░░░░░░░░░░░░░░░░░░░  ~200MB
OpenClaw      ████████████████████████████████████░░  ~500MB

Idle Memory (lower is better)

OpenIntentOS  █░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░  <15MB     ★
LangGraph     █████████████████░░░░░░░░░░░░░░░░░░░░░  ~180MB
CrewAI        ████████████████████░░░░░░░░░░░░░░░░░░  ~200MB
AutoGen       █████████████████████████░░░░░░░░░░░░░  ~250MB
OpenClaw      ████████████████████████████████████░░  ~394MB

Feature Comparison

Feature	OpenIntentOS	OpenClaw	CrewAI	AutoGen	LangGraph
Language	Rust 2024	TypeScript	Python	Python	Python
Binary size	~10 MB	~500 MB	~100 MB	~200 MB	~150 MB
Cold start	<200ms	~6s	~3s	~4s	~2.5s
Idle RAM	<15 MB	~394 MB	~200 MB	~250 MB	~180 MB
LLM cascade failover	7 providers	manual	manual	manual	manual
Encrypted vault	AES-256-GCM	config file	none	none	none
Wasm sandbox	wasmtime	none	none	Docker	none
Chat interfaces	Telegram · Discord · Feishu · WeChat · DingTalk · WeCom · QQ	Telegram · Discord	none	none	none
UI	CLI · TUI · Web · Desktop	Web only	none	none	none
SIMD intent router	aho-corasick	none	none	none	none
Autonomous schedules	cron adapter	cron	none	none	none
Local LLM (offline)	Ollama fallback	Ollama	Ollama	Ollama	Ollama
Zero-copy IPC	rkyv	JSON	none	none	none
Hot config reload	yes	no	no	no	no
License	MIT	MIT	MIT	Apache 2.0	MIT

Performance

Scenario	Target	How
Cold start	< 200ms	Single binary, no runtime, connection warmup
Simple intent	< 2µs	SIMD aho-corasick pattern match, no LLM call
SQLite read (hot)	< 0.05µs	moka lock-free cache
SQLite read (warm)	< 5µs	WAL + mmap 256 MiB
IPC message	< 1µs	rkyv zero-copy deserialization
Pattern matching	~2 GB/s	SIMD aho-corasick
Vector search	< 0.5ms	usearch, Rust native
Local classification	< 15ms	ONNX Runtime (ort)
Memory footprint	< 15 MB	Rust zero-overhead + mmap
Binary size	< 10 MB	LTO + strip + panic=abort

Architecture — Five-Layer Kernel

┌─────────────────────────────────────────────────────────────────┐
│  Layer 5 · Intent Interface                                     │
│  Telegram · Discord · Feishu · WeChat OA · DingTalk · WeCom     │
│  QQ Bot · REST · Cron · Natural language                        │
├─────────────────────────────────────────────────────────────────┤
│  Layer 4 · Agent Runtime  (tokio, async)                        │
│  Planner ── Executor ── Reflector ── LLM Router                 │
│            ReAct Loop  (zero-copy, same runtime)                │
├─────────────────────────────────────────────────────────────────┤
│  Layer 3 · Micro-Kernel                                         │
│  Scheduler (crossbeam)  Auth Vault (AES-256)  3-layer Memory    │
│  IPC Bus (rkyv)         Wasm Sandbox          ABAC Policy       │
│  Intent Router (SIMD)   Service Registry                        │
├─────────────────────────────────────────────────────────────────┤
│  Layer 2 · Adapters  (async trait, tokio)                       │
│  Filesystem  Shell  Web Search  Web Fetch  HTTP  Browser(CDP)   │
│  Email(IMAP) Calendar(CalDAV)  GitHub  Feishu  Slack  Cron      │
│  WeChat OA  DingTalk  WeCom  QQ Bot  Bilibili  Douyin  XHS      │
│  Weibo  Memory  Skills(WeCom·SMS·Weibo·Twitter·Email...)        │
├─────────────────────────────────────────────────────────────────┤
│  Layer 1 · Runtime & Storage                                    │
│  tokio · rusqlite WAL+mmap · moka cache · io_uring/kqueue       │
└─────────────────────────────────────────────────────────────────┘

Crate Graph

openintent-cli          (binary — single executable)
├── openintent-kernel   micro-kernel: scheduler, IPC bus, intent router
├── openintent-agent    ReAct loop, LLM client, tool dispatcher
│   └── openintent-kernel
├── openintent-store    SQLite WAL+mmap, 3-layer memory, user accounts
├── openintent-vault    AES-256-GCM encrypted secret storage
├── openintent-intent   intent classification + workflow engine
│   └── openintent-agent
├── openintent-adapters 20+ built-in adapters
│   └── openintent-vault
├── openintent-auth-engine  headless OAuth via CDP
│   └── openintent-vault
├── openintent-ui       iced GPU desktop UI
├── openintent-web      axum HTTP server + SSR
└── openintent-tui      ratatui terminal UI

Built-in Adapters

System & Productivity

Adapter	Capabilities
Filesystem	Read, write, list, search files. UTF-8-safe 16 KB truncation on large reads.
Shell	Execute commands with timeout, capture stdout/stderr. Full subprocess isolation.
Web Search	DuckDuckGo search with result ranking. No API key required.
Web Fetch	Full page fetch with text extraction. Handles JS-heavy sites via browser adapter.
HTTP Request	Arbitrary HTTP API calls — any method, headers, body.
Browser (CDP)	Chromium DevTools Protocol. Navigate, click, fill, screenshot. Headless OAuth flows.
Email (IMAP/SMTP)	Read inbox, send messages, manage folders. OAuth 2.0 with auto token refresh.
Calendar (CalDAV)	Create, read, update events. Works with Apple Calendar, Nextcloud, Google.
GitHub	List repos, read issues/PRs, post comments. Used for self-repair via evolution engine.
Cron	Schedule recurring tasks. Persistent across restarts via SQLite.
Memory	Working, episodic, and semantic memory layers. Vector search via usearch.

Messaging & Chat

Adapter	Capabilities	Auth
Telegram	Send messages, files, inline keyboards. Full bot API.	`TELEGRAM_BOT_TOKEN`
Discord	Send to channels, DMs, embeds.	`DISCORD_BOT_TOKEN`
Slack	Post to channels, DMs, threads.	`SLACK_BOT_TOKEN`
Feishu / Lark	Send to groups and DMs. Enterprise messaging.	`FEISHU_APP_*`
WeChat OA	Send text/image/news, get followers. Official Account API.	`WECHAT_OA_APP_*`
DingTalk	Webhook robot or enterprise app. Text, Markdown, action card.	`DINGTALK_*`
WeCom (企业微信)	Group robot or internal app. Text, Markdown, member list.	`WECOM_CORP_*`
QQ Official Bot	Channel messages, C2C direct messages, image push.	`QQ_BOT_APP_*`

Content & Social

Adapter	Capabilities	Auth
Douyin (抖音)	OAuth2, video upload & publish, video list & stats.	`DOUYIN_CLIENT_*`
XiaoHongShu (小红书)	Publish notes, search, stats, comments. HMAC-SHA256 signed.	`XHS_APP_*`
Weibo	OAuth2, post status, read mentions, reply.	`WEIBO_APP_*`
Bilibili Open Live	Live room info, send danmaku, danmaku history.	`BILI_ACCESS_KEY_*`

Skills (script-based, no Rust required)

Skill	Capabilities	Auth
wecom-bot	WeCom group robot: text, Markdown, news card.	`WECOM_WEBHOOK_URL`
sms-tencent	Tencent Cloud SMS, TC3-HMAC-SHA256 signed.	`TENCENTCLOUD_*`
sms-aliyun	Alibaba Cloud SMS, HMAC-SHA1 signed.	`ALIYUN_*`
weibo-post	Weibo post/mentions/reply via shell.	`WEIBO_ACCESS_TOKEN`
twitter-manager	Post tweets and manage Twitter via OAuth1.	`TWITTER_*`
email-automation	Send email via SMTP with Python.	SMTP env vars
daily-briefing	Morning digest: weather + news + calendar.	configurable
web-search-plus	Enhanced web search via shell.	none
weather-check	Weather lookup by city.	`OPENWEATHER_API_KEY`
ip-lookup	Geo-locate an IP address.	none

Bot Commands

Command	Description
`/models`	Inline keyboard of all available model aliases
`/model <alias>`	Switch LLM for this chat — persists across restarts
`/tokens on\|off`	Toggle token usage display after each response
`/status`	Current provider, model, and health
`/clear`	Clear conversation history for this chat
`/reset`	Restore default provider and model
`/help`	List all commands

Model Aliases

Alias	Provider	Model	Free?
`claude`	Anthropic	claude-sonnet-4-6
`gpt4o`	OpenAI	gpt-4o
`deepseek`	DeepSeek	deepseek-chat	✓ credit
`siliconflow` / `sf`	SiliconFlow 硅基流动	DeepSeek-V3	✓ 14M/mo 🇨🇳
`sf-qwen`	SiliconFlow	Qwen3-235B	✓ 🇨🇳
`moonshot` / `kimi-cn`	Moonshot Kimi	moonshot-v1-8k	✓ credit 🇨🇳
`glm` / `glm-flash`	Zhipu GLM	glm-4-flash	✓ free 🇨🇳
`tongyi`	Tongyi Qwen	qwen-turbo	✓ tier 🇨🇳
`nvidia`	NVIDIA NIM	qwen/qwen3.5-397b-a17b	✓ 1000/mo
`gemini`	Google	gemini-2.5-flash	✓ tier
`groq`	Groq	llama-3.3-70b-versatile	✓ free
`grok`	xAI	grok-3
`mistral`	Mistral	mistral-large
`ollama` / `local`	Ollama	qwen2.5	✓ local

Security

Layer	Mechanism
Credential storage	AES-256-GCM encrypted vault. Master key in macOS Keychain / Linux Secret Service.
Transport	TLS 1.2+ enforced on all outbound connections via reqwest + rustls.
Wasm sandbox	Third-party plugin code runs in wasmtime with fuel metering.
Path traversal	Filesystem adapter canonicalises paths and rejects symlink escapes.
Secret in memory	API keys held in locked memory and zeroed after use.
OAuth flows	PKCE + state parameter on all OAuth 2.0 flows. Tokens stored encrypted.
Audit log	Append-only log of every external action (SQLite).
No telemetry	Fully self-hosted. Zero data leaves your machine except explicit LLM API calls.

Quick Start

One-line install (recommended)

macOS / Linux / WSL / Raspberry Pi / Termux:

curl -fsSL https://raw.githubusercontent.com/OpenIntentOS/OpenIntentOS/main/install.sh | bash

Windows 10 / 11 — open PowerShell as normal user (no admin needed):

irm https://raw.githubusercontent.com/OpenIntentOS/OpenIntentOS/main/install.ps1 | iex

The installer will guide you through everything step by step.

What the installer does

Step 1/5 · Detecting your system
  ✓  macOS (Apple Silicon)

Step 2/5 · Downloading OpenIntentOS
  ✓  Downloaded 9.8 MB binary (v0.1.0)

Step 3/5 · Connect your AI providers
  📱 Telegram Bot Token
     Don't have one? Open Telegram → @BotFather → /newbot
  Enter: ████████████████████████

  🧠 OpenAI API Key  (https://platform.openai.com/api-keys)
  Enter: ████████████████████████

  🧠 NVIDIA NIM Key  (free tier — https://build.nvidia.com)
  Enter: (skipped)

Step 4/5 · Installing system service
  ✓  macOS LaunchAgent installed (auto-starts on login)

Step 5/5 · Verifying bot is running
  ✓  Bot is running

  ✓  OpenIntentOS is installed and running!
     Open Telegram and message your bot to get started.

After install — useful commands

~/.openintentos/status.sh     # check if bot is running + last 20 log lines
~/.openintentos/restart.sh    # apply config changes without reinstalling
~/.openintentos/uninstall.sh  # remove everything cleanly
tail -f ~/.openintentos/bot.log  # live log stream

Update

# Run the installer again — existing API keys and data are preserved
curl -fsSL https://raw.githubusercontent.com/OpenIntentOS/OpenIntentOS/main/install.sh | bash

Build from source (developers)

git clone https://github.com/OpenIntentOS/OpenIntentOS
cd OpenIntentOS
cargo build --release
# Binary: ./target/release/openintent-cli
export TELEGRAM_BOT_TOKEN="..." OPENAI_API_KEY="..."
./target/release/openintent serve

Configuration

All runtime behaviour lives in config/default.toml. Hot-reloaded on file change — no restart required.

[provider]
name     = "openai"
base_url = "https://api.openai.com/v1"
model    = "chatgpt-pro"

[bot]
history_window         = 20     # messages of context per chat
show_token_usage       = false  # toggle with /tokens on|off
simple_query_threshold = 120    # chars; short messages use cheap model

[agent]
max_react_turns   = 10
tool_timeout_secs = 30

Full reference: Configuration Docs

Tech Stack

Component	Technology	Why
Language	Rust 2024	Zero-cost abstractions, no GC, no runtime
Async	tokio	io_uring (Linux), kqueue (macOS)
HTTP client	reqwest	HTTP/2, connection pooling, rustls
Database	rusqlite WAL+mmap	<5µs reads, embedded, no daemon
Hot cache	moka	Lock-free concurrent, no mutex contention
Vector search	usearch	<0.5ms, Rust native, SIMD
Local inference	ort (ONNX Runtime)	<15ms classification
String matching	aho-corasick (SIMD)	~2 GB/s, zero alloc
Serialisation	rkyv	Zero-copy deserialization
JSON	simd-json + serde	SIMD-accelerated parsing
Encryption	ring	AES-256-GCM, hardware AES-NI
Desktop UI	iced	Pure Rust GPU-rendered
Web UI	axum + askama	Lightweight SSR, no JS framework
TUI	ratatui	Terminal UI for headless servers
Browser	chromiumoxide	Rust-native CDP client
Wasm sandbox	wasmtime	Fuel metering, secure isolation
Error types	thiserror	Library errors; anyhow in CLI only
Logging	tracing	Structured, async-aware, zero-cost spans

Documentation

Getting Started — build, configure, run
Architecture — five-layer design, crate graph, failover internals
Configuration Reference — every config key documented

License

MIT

Name		Name	Last commit message	Last commit date
Latest commit History 117 Commits
.github/workflows		.github/workflows
assets		assets
config		config
crates		crates
data		data
deploy		deploy
docs		docs
skills		skills
.env.example		.env.example
.gitignore		.gitignore
CHANGELOG.md		CHANGELOG.md
CLAUDE.md		CLAUDE.md
Cargo.lock		Cargo.lock
Cargo.toml		Cargo.toml
Dockerfile		Dockerfile
Makefile		Makefile
PLAN.md		PLAN.md
README.md		README.md
docker-compose.yml		docker-compose.yml
install.ps1		install.ps1
install.sh		install.sh
rust-toolchain.toml		rust-toolchain.toml
test_e2e_local.sh		test_e2e_local.sh

Folders and files

Latest commit

History

Repository files navigation

OpenIntentOS

The Intent-Driven AI Operating System

Install

No API Key? Start for Free

🇨🇳 China — No VPN Required

🌍 International

Run Locally (No Internet)

What is OpenIntentOS?

What it looks like

Cascading LLM Failover — Seven Providers, Zero Downtime

OpenIntentOS vs The Landscape

Cold Start Time (lower is better)

Binary / Install Size (lower is better)

Idle Memory (lower is better)

Feature Comparison

Performance

Architecture — Five-Layer Kernel

Crate Graph

Built-in Adapters

System & Productivity

Messaging & Chat

Content & Social

Skills (script-based, no Rust required)

Bot Commands

Model Aliases

Security

Quick Start

One-line install (recommended)

What the installer does

After install — useful commands

Update

Build from source (developers)

Configuration

Tech Stack

Documentation

License

About

Resources

Uh oh!

Stars

Watchers

Forks

Releases 12

Packages 0

Uh oh!

Contributors

Uh oh!

Languages

Packages