Agent systems become difficult to trust when their most important choices are hidden inside control flow. Platonic begins from a stricter premise: work that changes an outcome should cross a typed boundary and leave durable evidence.

The server

Platonic owns workspaces, agents, threads, provider calls, tools, policy, approvals, and ledgers. It assembles bounded context, separates tool proposals from policy decisions, and records replay and audit truth in the event ledger.

The client distribution

Plato Agent is the client distribution built on Platonic. It packages the CLI, TUI, and desktop clients with curated agent configurations and skills; server execution stays in Platonic.

A thread can continue across clients, pause for a local approval, recover after interruption, and replay from recorded events. The clients make those server boundaries visible without owning them.

Plato Agent TUI showing a completed read-only task and its recorded result.
Plato Agent is the client distribution; Platonic is the server.

Start with evidence

The quickest way to understand the system is to run a bounded task and replay its ledger. The tagged quickstart covers the supported bundle, first run, approvals, and offline readback without creating a second documentation path here.

Open the 0.2.2 quickstart Explore the source