Atom
Self-hosted AI agent platform featuring built-in governance, policy checks, and durable human approvals.
Train self-hosted AI workers with custom rules, explicit governance, and reviewable document execution.
Atom is a self-hosted AI agent platform designed for businesses that need strict oversight over automated tasks. You can define custom rules, field mappings, and human approval checkpoints before agents execute actions or publish deliverables. It suits operations teams and technical leaders who want governance controls built directly into agent execution paths.
What it is
Atom processes business tasks like sales quoting, invoice checking, and document generation by executing governed agentic workflows. You teach agents reusable business rules, link data sources, and review source calculations before final outputs are generated. Long-running tasks survive server restarts, holding uncertain actions until human approval is recorded.
Who it's for
This platform suits technical business managers and developers who want local control over automated context and data. It is not intended for non-technical users seeking a plug-and-play cloud assistant without infrastructure management. If your team cannot maintain a Python and Node.js environment, this project is not for you.
Key features
Human in the loop controls
Governed actions pause for human approval, preventing mismatched or unverified model decisions from running.
Reusable contextual teaching snapshots
Saved field mappings and rules apply to new assignments without altering active jobs already in progress.
Durable multi-hour execution state
Long-running tasks and pending approvals survive backend process restarts without losing operational history.
Flexible model routing endpoints
Connect external model providers or local OpenAI-compatible endpoints directly to your single-tenant workspace.
Up and running in 3 steps
Our honest verdict
What's great
- Deterministic rule policy checks prevent hallucinated approvals from completing execution
- Source-grounded provenance maps calculations directly to generated Word and PowerPoint files
- Active jobs retain baseline rules when global teaching instructions update
- Local sample workspace runs out of the box without requiring model API keys
Where it falls short
- Requires managing local Python 3.11–3.13 and Node.js environments
- General autonomous judgment and broad business automation remain ongoing development goals
- Underlying model API fees apply when connecting hosted LLM providers
- Single-tenant architecture requires manual server maintenance and local security handling
Pricing & license
Atom is free under the AGPL-3.0 license. There are no paid hosted tiers provided by the repository; your expenses depend on local hosting infrastructure and external model API usage.
Open remote AI-training roles
873 open remote AI-training roles, updated November 15, 2026; most pay $20–$170 an hour. Zekai lists them free; the application goes to the employer’s platform.
Alternatives
These reviewed AI tools do the same job instantly — from our scored directory.
Use it, or have us set it up
FAQ
Does Atom require an internet connection or paid cloud LLM key to test?
No, the starter local sample runs locally without cloud credentials, external model APIs, or active network requests.
What happens if an agent encounters an uncertain rule or calculation?
Atom pauses execution at policy checkpoints, retaining unresolved requirements until a human user reconciles the action.
