Give every employee a team of AI agents — trusted and safe by design.
Atom is an open-source, self-hosted AI agent workforce for your employees. Instead of one assistant, Atom runs a team of specialty agents — sales, support, finance, engineering — that your people delegate to in plain language. Agents plan, verify, and execute complex workflows across your entire tech stack.
Autonomous agents as teammates: Atom's agents don't just respond to commands — they operate autonomously within governed boundaries, handling routine work end-to-end while your employees focus on strategy, creativity, and high-value decisions. Each agent is a digital teammate with specialized skills, persistent memory, and the ability to collaborate with other agents and humans alike.
Trusted by design: every agent action is governed by a 4-tier maturity model, executed inside a default-on sandbox, and recorded in a complete audit trail — with human-in-the-loop approval wherever you want it. Your employees get capable help; you keep control.
Your data stays yours: workflow data, agent state, and memory live on your infrastructure. LLM inference uses your own API keys (BYOK) — or local models (Ollama/Llama.cpp) for fully private deployments.
💰 Budget-friendly AI agents: OpenCode Go subscription (~90% savings vs pay-per-token) — one $10/mo key unlocks general-purpose models (DeepSeek V4, Kimi K3, GLM 5.2, MiniMax M3, Qwen 3.7, Nemotron 3 Ultra, Grok 4.5) with full tool-calling & structured output — not just for coding, works for any agent workload. Setup guide →
No lock-in: 16+ LLM providers (OpenAI, Anthropic, DeepSeek, Gemini, MiniMax, Groq…) with automatic cost-aware routing, fallback, and self-healing.
Describe the outcome. Atom builds and runs the workflow.
| What you say | What Atom delivers |
|---|---|
| "When a lead comes in HubSpot, research the company, score it, create an Asana task for the rep, and ping Slack" | A governed, replayable workflow with Human-in-the-Loop approval gates |
| "Extract invoice data from Gmail PDFs, match against QuickBooks, flag discrepancies" | End-to-end pipeline: Gmail → PDF OCR → QuickBooks reconciliation → Slack alert |
| "Monitor Zendesk tickets for sentiment, auto-escalate urgent ones, draft replies" | Real-time triage agent with approval before send |
| "Generate a weekly sales report from Salesforce, format in Excel, email the team" | Scheduled workflow: SOQL query → formula-evaluated Excel → Office 365 send |
Why it's different from Zapier/Make/n8n:
- Agents, not just steps — Agents reason (not just execute): they research, decide, retry, and self-correct
- Governance built-in — Maturity gates, HITL approval, audit trail, sandbox isolation
- Self-hosted & private — Your data, your keys, your infrastructure
- Office-native — Real Excel/Word/PPTX with formula evaluation, live Canvas co-editing
- Agent-authored — You can chat with an agent to build/modify workflows (no drag-and-drop required)
Workflow Automation Guide → · Quick Start →
┌─────────────────────────────────────────────────────┐
│ AI AGENT SPECTRUM │
├─────────────────┬─────────────────┬─────────────────┤
│ CONSUMER │ DEVELOPER │ ENTERPRISE │
│ ASSISTANTS │ FRAMEWORKS │ WORKFORCE │
├─────────────────┼─────────────────┼─────────────────┤
│ ChatGPT/Claude │ LangGraph │ ✅ ATOM │
│ Notion AI │ AutoGPT │ │
│ Copilot │ CrewAI │ │
│ Perplexity │ AutoGen │ │
├─────────────────┼─────────────────┼─────────────────┤
│ Single chat │ Code-first │ Team delegation │
│ Reactive only │ Build-your-own │ Governed by │
│ Cloud-only │ Self-hosted │ design │
│ No integrations │ Bring integrations│ 46+ native │
└─────────────────┴─────────────────┴─────────────────┘
Atom is the only open-source platform that delivers:
- Enterprise governance (maturity tiers, HITL, audit) — without vendor lock-in
- Self-hosted privacy — your data, your keys, your infrastructure
- Autonomous agent teammates — agents that work with your people, not just for them
- Agent-authored workflows — chat to build, no drag-and-drop
- Office/Canvas native — real Excel formulas, live co-editing
- 46+ business integrations — Salesforce, HubSpot, Slack, Jira, Stripe, QuickBooks…
Atom redefines the relationship between humans and AI in the workplace. Instead of a single chat assistant, your employees get a team of autonomous digital teammates:
| Traditional AI Assistant | Atom Autonomous Agent Teammates |
|---|---|
| Reactive — waits for commands | Proactive — handles routine work end-to-end |
| Single-threaded, no memory | Persistent memory, cross-session continuity |
| No governance, no audit | 4-tier maturity, HITL approval, full audit trail |
| One generic model per task | Specialized agents per domain (sales, finance, support, eng) |
| Human does the orchestration | Agents collaborate with each other & humans |
| Cloud-only, data leaves your infra | Self-hosted, your keys, your infrastructure |
The vision: Every employee gets a personal agent team that knows their workflows, remembers context across days/weeks, and autonomously executes the repetitive 60-80% of work — research, data entry, scheduling, drafting, reconciliation — so people can focus on judgment, creativity, and relationships.
This isn't "AI replacing humans." It's AI handling the work humans shouldn't be doing, with governance that keeps you in control.
| Alternative | Focus | Key Difference | Deep Dive |
|---|---|---|---|
| Hermes Agent (Nous Research) | Personal coding/productivity assistant | Single-agent, no governance, no integrations, no sandbox | Atom vs Hermes → |
| OpenClaw | Personal productivity, messaging-first | Single-agent, Markdown memory, smart home focus | Atom vs OpenClaw → |
| LangGraph / CrewAI / AutoGen | Developer frameworks | Code-first, build-your-own governance & integrations | Why Atom? |
| Zapier / Make / n8n | Workflow automation | Step-based (not agents), no reasoning, no governance | AI-Generated Workflow Automation |
TL;DR: If you're evaluating personal agents → Hermes/OpenClaw. If you need governed multi-agent business automation → Atom.
| Category | Features |
|---|---|
| 🤖 Multi-Agent Orchestration | Queen Agent (structured workflows) + Fleet Admiral (open-ended tasks) + Conductor (5 execution strategies) + validated state machine with rollback; governed fleet routing with ranked specialist matching |
| 🛡️ Governance & Safety | 4-tier maturity (Student→Autonomous), 3-layer policy engine, HITL approval, complete audit trail, AI-powered training |
| ✅ Outcome Verification | Postcondition oracle re-derives success against the system of record (no self-attestation), two-tier confidence provenance, reviewer re-delegation loop with diversity-aware sampling |
| 🧠 Memory & Learning | Per-turn fact extraction, 2-tier recall (SQL + LanceDB), episodic memory, memory_remember/forget, self-evolution (Memento/AlphaEvolver, self-evolving harness) |
| 🔎 Hybrid Search | documents.search fuses BM25 (FTS5/tsvector) + vector (LanceDB) via Reciprocal Rank Fusion (RRF) — semantic + precise retrieval with citations |
| 🗂️ Knowledge VFS | Agent-native document tree — ls/cat/grep/search with line-numbered citations instead of bespoke per-store queries |
| 📻 Agent Radio | Lateral peer-to-peer messaging between agents (mention-first, budget-governed) — agents coordinate without hardcoded teams |
| 💼 Office Automation | Real-time Excel/Word/PPTX co-editing on Canvas; formula-evaluating workbook runtime; agent↔document sync |
| 🧩 Mini-Apps | Agent-authored stateful canvas apps (spreadsheets/docs/decks) — Firecracker microVM isolation, per-instance chat co-editing |
| 🔍 GraphRAG & Intelligence | Multi-hop expansion, Leiden community detection, JIT fact verification, D3 visual explorer |
| 🌐 46+ Business Integrations | Salesforce, HubSpot, Slack, Teams, Gmail, Notion, Jira, Linear, Stripe, QuickBooks, Shopify, GitHub, GitLab, Zoom… |
| 🛰️ LLM Gateway | OpenAI/Anthropic-compatible API over your BYOK — point Claude Code, n8n, or any OpenAI-SDK app at Atom |
| 💰 Cost-Aware Routing | 5-tier cognitive classification, 16+ providers, learning router (feedback-based re-ranking), RTK token compression |
| 🎯 Goal-Driven Loops | Agents terminate on a definition_of_done predicate instead of always burning to max_steps; utility targets, custom action surfaces, stuck-detection |
| Layer | What you get |
|---|---|
| Execution Sandbox | Filesystem scope, tool whitelist, tripwires, resource caps, KillRun — enforced on every dispatch path |
| Encrypted Credentials | OAuth integration tokens encrypted at rest (Fernet); production fails closed without key |
| Per-Agent Capability Bindings | Zero-trust tool scoping — agent can never exceed its tier floor |
| Outbound Gatekeeper | Rate limiting, response masking, HITL mutation approval on integration calls |
| Data-Taint Tracking | Restricted data observed in a run blocks external outbound actions |
| External MCP Client | Connect to arbitrary external MCP servers (Cloudflare portals) |
Security Architecture → · Sandbox Deep-Dive →
git clone https://github.com/rush86999/atom.git && cd atom
make setup # one-shot dev bootstrap (venv, deps, .env, frontend)
make backend # full backend on :8001
# in a second terminal:
make frontend # Next.js UI on :3001Then set one LLM key in backend/.env:
OPENAI_API_KEY/ANTHROPIC_API_KEY/DEEPSEEK_API_KEY/GOOGLE_API_KEY… orOPENCODE_API_KEYfor low-cost subscription coding models (~90% savings) orATOM_LOCAL_ONLY=true+OLLAMA_BASE_URL=http://localhost:11434/v1for fully local
Open http://localhost:3001 — Sign in as admin@example.com (password in backend/logs/bootstrap_admin_password.txt)
Full Quick Start → · Docker → · DigitalOcean 1-Click →
| Department | Scenario | Key Integrations |
|---|---|---|
| Sales | New HubSpot lead → Research company → Score → Asana task → Slack notify | HubSpot, Asana, Slack, LinkedIn |
| Finance | Gmail PDF invoice → OCR extract → QuickBooks match → Flag discrepancies | Gmail, QuickBooks, Excel, Slack |
| Support | Zendesk ticket → Sentiment analysis → Auto-escalate urgent → Draft reply | Zendesk, Slack, Email |
| HR | BambooHR new hire → Provision accounts → Invite to Slack → Schedule orientation | BambooHR, Google Workspace, Slack, Calendar |
| Engineering | GitHub PR → Run tests → Security scan → Post summary → Auto-merge if green | GitHub, GitLab, Slack, Jira |
| Marketing | Content calendar → Generate posts → Human review → Schedule multi-platform | Notion, Slack, LinkedIn, Twitter, Meta |
atom/
├── backend/ # FastAPI app — main_api_app:app (full) / minimal_app:app (smoke)
├── frontend-nextjs/ # Next.js web UI
├── mobile/ # React Native (Expo) companion app
├── menubar/ # Tauri macOS menubar companion
├── scripts/ · infra/ · installer/ · examples/
├── docs/ # project documentation
├── Dockerfile # dual-app image (backend + frontend)
└── Makefile # common tasks (start here)
| Atom | Zapier/Make/n8n | LangGraph/CrewAI | OpenClaw | LangChain | |
|---|---|---|---|---|---|
| AI Agents (reason, not just execute) | ✅ | ❌ | ✅ | ✅ | ✅ |
| Governance (maturity + HITL + audit) | ✅ | ❌ | ❌ | ❌ | ❌ |
| Default-on Sandbox (all dispatch paths) | ✅ | ❌ | ❌ | ❌ | ❌ |
| Self-Hosted / Private (your keys, your infra) | ✅ | ❌ | ✅ | ✅ | ✅ |
| Office/Canvas Native (Excel formulas, co-edit) | ✅ | ❌ | ❌ | ❌ | ❌ |
| Agent-Authored Workflows (chat to build) | ✅ | ❌ | ❌ | ❌ | ❌ |
| 46+ Business Integrations (CRM, finance, support) | ✅ | ✅ | — | 50+ personal | — |
| Cost-Aware LLM Routing (16+ providers) | ✅ | ❌ | ◐ | ◐ | ◐ |
| Mini-Apps (agent-authored stateful apps) | ✅ | ❌ | ❌ | ❌ | ❌ |
| GraphRAG / Episodic Memory | ✅ | ❌ | ◐ | ◐ | ◐ |
We welcome contributions — see CONTRIBUTING.md. Quality bar: tests pass (100%), coverage ≥70%, review required, docs updated.
- Issues: GitHub Issues
- Blog: Substack
- License: AGPL v3 — LICENSE.md
Built with FastAPI | SQLAlchemy | LangChain | Playwright | Next.js
Experience the future of self-hosted AI automation — safe enough for your whole team.
⭐ Star us on GitHub — it helps!