Built on Molecule OS and Platform Services

The AI-native shared inbox

Build one AI agent and run it across email, chat, and social — grounded in your company's knowledge, governed by one control plane.

The platform

Three systems, one platform

Atoms is the product. Molecule OS builds and runs the agents. Platform Services governs every tenant.

The product

The AI inbox where a company authors an agent and invokes runs. It never stores or runs agents — Molecule does.

The agent runtime

Holds an agent's definition and runs it as a typed graph, grounded in the company's knowledge. The brain is one part of it.

The control plane

One company id, entitlements, WorkOS identity, and the signed config snapshots the runtime caches locally and reads off the hot path.

Under the hood

Atoms invokes a run. Molecule walks the agent's graph and streams the outcome back. One path, whether the answer is a reply or an action.

1
Atoms invokes a run
Invoke
The only UI. It authors the agent and calls Molecule to run it — it never stores the definition or walks the graph itself.
2
Molecule runs the agent
Graph walk
The runtime interprets the agent's typed graph. Retrieval is its most-used step, not a parallel path; external tools are called through the broker, which holds the credential so the model never sees a token.
Assemble scopeRefine and routeCall one toolValidateRespond or continue
3
The outcome streams back
SSE
Tokens, tool events, and the finished outcome — send, draft, escalate, or a tool call — stream to Atoms over SSE, and Atoms renders the run as it happens.
Platform Services supplies a signed config snapshot — policy, entitlements, identity, and the connection registry — cached locally and read on every step, never a synchronous call per request.

The documentation, the API reference, the roadmap, and the dashboard all live inside the workspace.