The Operating System for Personal AI
Website · Docs · Marketplace · Discord
ArgentOS is a personal AI operating system that runs on your hardware, speaks through your channels, remembers your context, and works autonomously on your behalf. Core is the free, open-source foundation.
Prerequisite: Homebrew (macOS) — the installer uses it for PostgreSQL, Redis, and system services.
curl -fsSL https://argentos.ai/install.sh | bashmacOS first. The installer handles Node, pnpm, PostgreSQL, Redis, the local dashboard runtime, and onboarding/browser handoff automatically.
Or from source:
git clone https://github.com/ArgentAIOS/argentos-core.git
cd argentos-core
pnpm install && pnpm build
bash install.sh| System | What It Does |
|---|---|
| Memory | SQLite + pgvector hybrid search, auto-capture, entity tracking, embeddings |
| Heartbeat | Periodic accountability, contract/promise system, scoring |
| Contemplation | Always-on thinking loop, local models via Ollama/LM Studio, episode journaling |
| Channels | Telegram, Discord, Slack, WhatsApp, Signal, iMessage, and more |
| Workflows | Visual multi-agent pipeline builder with 62 connector integrations |
| Dashboard | React + WebGL avatar, chat, task board, alignment docs editor |
| 62 Connectors | Stripe, HubSpot, GitHub, Notion, Salesforce, Jira, Twilio, and dozens more |
| Intent System | Three-tier hierarchical policy engine for safe agent behavior |
| Model Router | Complexity scoring with local/fast/balanced/powerful tier routing |
| SIS | Self-improving system — lesson extraction from experience |
| Knowledge Library | RAG with agent-level ACL, PDF/DOCX/XLSX ingestion |
| Encrypted Secrets | AES-256-GCM credential storage with OS keychain integration |
| Tasks & Projects | Full CRUD with priorities, dependencies, FTS |
| Alignment Docs | SOUL.md, IDENTITY.md, USER.md — define who your agent is |
| Backup | SQLite backup with local/S3/R2 upload, compression, restore |
The ArgentOS Marketplace has 116+ packages:
argent marketplace install <package>Browse connectors, skills, plugins, and workflow templates. Every package is scanned with VirusTotal.
src/
├── memory/ # Memo + MemU — hybrid search, embeddings, entity tracking
├── infra/ # Contemplation, heartbeat, SIS, workflows, backup
├── agents/ # Agent runtime, tools, intent system, model router
├── channels/ # Telegram, Discord, Slack, WhatsApp, and more
├── data/ # Storage adapters (SQLite, PostgreSQL, Redis)
├── gateway/ # WebSocket control plane, server methods
├── models/ # Model router, provider registry
└── cli/ # CLI commands
dashboard/ # React dashboard (36K+ LOC)
tools/aos/ # 62 connector harnesses
apps/ # macOS, iOS, Android apps
pnpm install
pnpm build
pnpm testWe want Core to be community-driven:
- Star this repo — it helps more than you think
- Join Discord — this is where the community lives
- File issues — bugs, feature requests, questions
- Submit PRs — we review everything and good contributions ship fast
- Build connectors — the pattern is documented, each one helps the whole ecosystem
TypeScript (ESM), strict typing, Oxlint + Oxfmt. See CONTRIBUTING.md for details.
Core is the free foundation. ArgentOS Business adds:
- Multi-agent workforce management
- Execution worker with approval workflows
- Job orchestrator for team scheduling
- Industry-specific intent packs (legal, healthcare, MSP, finance)
- Operations dashboard (Workflow Map, Org Chart, Workloads)
| Website | argentos.ai |
| Docs | docs.argent.ai |
| Marketplace | marketplace.argentos.ai |
| Discord | discord.gg/argentos |
| X | @ArgentAIOS |
ArgentOS stands on the work of others:
- pi-mono by Mario Zechner (@badlogic) — the agent runtime core that ArgentOS builds upon. pi-mono provides the foundational agent loop, session management, and tool execution framework.
- OpenClaw by Peter Steinberger — the original fork that gave ArgentOS its launching pad. OpenClaw's architecture informed early design decisions for the gateway, plugin system, and channel infrastructure.
- TOON — Token-Oriented Object Notation. ArgentOS uses TOON throughout the system to encode structured data for LLM prompts at 40-50% fewer tokens than JSON. Used in workflow pipeline context, agent handoffs, memory recall, tool results, task breakdowns, and team status.
MIT License. See LICENSE for details.
An ArgentOS project by Jason Brashear · Deployed on Railway · Powered by CodeRabbit & Blacksmith · Open source (MIT)
Built in Texas. Runs on your hardware. Answers to you.
