Governed agentic operating system for orchestration, evidence, mutations, engineering intelligence, and multi-channel runtime operations.
AethOS separates orchestration from execution through policy, evidence, verification, approval, and audit.
Install · Architecture · Mission Control · Contribute · Docs
AethOS is a secure operational intelligence platform — not a chatbot wrapper and not an unrestricted shell.
It routes every operational action through a governed lifecycle:
Channels → Orchestration → Policy → Provider Runtime → Evidence → Verification → Mission Control → Audit
Why it matters: multi-channel intents (chat, Telegram, Mission Control) converge on one orchestration brain with readonly-first analysis, explicit approval for mutations, and auditable artifacts.
Why trust it: no hidden automation · no auto-merge · no silent background coding · evidence before action.
Operational orchestration runtime for governed AI systems.
macOS or Linux:
curl -fsSL https://raw.githubusercontent.com/pilotmain/AethOS/main/install.sh | bashWindows PowerShell:
irm https://raw.githubusercontent.com/pilotmain/AethOS/main/install.ps1 | iexWindows with curl.exe:
curl.exe -fsSL https://raw.githubusercontent.com/pilotmain/AethOS/main/install.ps1 | powershell.exe -NoProfile -ExecutionPolicy Bypass -Command -The install finishes with an interactive setup wizard (it works inside the
one-line curl flow): pick your AI provider — Anthropic, OpenRouter, OpenAI,
Gemini, Mistral, Groq, xAI, DeepSeek, Together, or a local Ollama/LM Studio
server — and your key is validated live and written to .env, along with
self-host mode, a generated vault key, and an optional login passphrase. When
it ends, the first run is ready to use. Re-run it any time with
./install.sh --onboard; already-installed machines get an
update / reinstall / onboard menu, plus --update, --reinstall, and
--uninstall.
The installer is idempotent and checkpoints every stage. If a network, dependency, or verification step stops, fix the reported cause and continue:
./install.sh --resume # macOS / Linux.\install.ps1 -Resume # WindowsEvery stage has focused help (./install.sh --help-step preflight or
.\install.ps1 -HelpStep preflight) and a status view (--status / -Status).
See the complete installation guide for prerequisites,
custom paths, API-only installs, recovery, and upgrades.
Or install from a local clone:
git clone https://github.com/pilotmain/AethOS.git
cd AethOS
./install.shThe installer will:
- Preserve a local checkout or clone into
~/aethos(override withAETHOS_INSTALL_DIR) - Checkpoint preflight, source, backend, frontend, and verification stages
- Install the Mission Control API runtime (FastAPI · port
8010) - Configure the local orchestration environment (
.envfrom template) - Install cloud adapters and secure local-vault support
- Set up the Mission Control web UI (Next.js · port
3000) - Leave mutation execution disabled by default — enable explicitly in
.env - Verify the API, CLI, and frontend toolchain before reporting success
- Print the first-run URL, doctor command, recovery command, and log path
Then launch:
./run.shOn Windows use .\run.ps1. Both launchers support --help / -Help and
API-only or web-only modes.
Open the first-run URL below. You'll sign in (if a login passphrase is set), then a short first-run setup learns your name, working hours, timezone, and preferred tone so AethOS can address you naturally.
| Surface | URL |
|---|---|
| First-run / Mission Control UI | http://localhost:3000 |
| Mission Control API | http://127.0.0.1:8010 |
| Health | http://127.0.0.1:8010/api/v1/health |
AethOS uses a model-as-brain pattern: the Mission Control provider (Anthropic, OpenRouter, local LLM, etc.) routes and answers; AethOS exposes tools, executes governed operations, and enforces policy. Capabilities are provider-agnostic — generic over the MC provider registry and per-tenant credential vault, not hardcoded to any single cloud.
Channels (Web · Telegram · API)
│
▼
Single-loop orchestration
(model tool loop · readonly-first routing)
│
▼
Policy + Governance Layer
(approval · blast radius · blocks)
│
▼
Provider runtime (registry + vault)
Railway · GitHub · Vercel · AWS · …
│
▼
Evidence + Verification Engine
(browser capture · artifacts · tests)
│
▼
Mission Control
(observability · audit · engineering)
Reads run directly. Mutations route through preflight → human approval → the correct provider tool.
These are the capabilities verified by pytest tests -q -k beta_smoke and advertised in onboarding:
- One model-driven chat loop with governed mutation gates — questions never become deploy targets
- Summarize URLs and follow up from conversation memory without canned help blurbs
- Render structured operational views to the live Canvas panel
- Send Telegram messages with real transport errors surfaced (no generic failure literals)
- Diagnose channel health (Telegram webhook, tokens) before workspace or repo commands
- Run Railway read-only deployment and inventory checks directly — no preflight job
- List provider inventory with live health tables across every connected cloud
- Analyze connected GitHub repositories via API — no local workspace registration required
- Run multi-model arbiter consensus when at least two models are configured
- Check deploy readiness directly; route deploy mutations through approval with the right provider tool
- Review connected repo structure, dependencies, and CI workflows via GitHub API
Multi-tenant isolation and mutation governance apply to every capability above.
Mission Control is the operational surface — designed for orchestration clarity, not chat aesthetics.
- Sidebar navigation: Runtime · Providers · Browser · Engineering · System
- Provider cards with collapsed credential management
- Mutation audit trail and operation preflights
- Engineering workspace cards, architecture maps, dependency health
Open after ./run.sh: http://localhost:3000/mission-control
| Guarantee | Status |
|---|---|
| Readonly-first engineering analysis | Enforced |
| Mutation preflight before writes | Enforced |
| Human approval for governed execution | Required |
| Browser capture policy tiers | Enforced |
| Audit artifacts for evidence operations | Generated |
| Unrestricted shell / auto-merge | Blocked |
git clone https://github.com/pilotmain/AethOS.git
cd AethOS
cp .env.example .env
python3 -m venv .venv && source .venv/bin/activate
pip install -e ".[dev]"
uvicorn aethos_core.api.main:app --reload --port 8010cd web
echo 'NEXT_PUBLIC_API_BASE=http://127.0.0.1:8010' > .env.local
npm install && npm run devSee docs/SETUP.md for provider keys, browser automation, and Telegram.
| Doc | Topic |
|---|---|
| Documentation index | Supported user, operator, and developer guides |
| Installation | One-command install, resume, Windows, upgrades |
| Setup | Full install and environment |
| Brand & install experience | Visual identity, terminology, onboarding |
| Architecture | System design |
| Runtime | Chat lanes, jobs, orchestration |
| Testing | Test strategy |
pytest tests/ -q
cd web && npm test -- --runReliability gate: pytest tests/test_chat_20_turn.py -q
Before opening a pull request, read CONTRIBUTING.md. All contributions require review, passing checks, and a DCO sign-off. Project roles and decisions are described in GOVERNANCE.md.
AethOS is open source under the Apache License 2.0. See LICENSING.md and COPYRIGHT.md for scope, attribution, and contributor ownership.
AethOS — confidence at first command.