Skip to content

Use Plato Agent

User docs

Unreleased 0.2.0 documentation: This page describes the current develop behavior. The public release is still 0.1.0; no released 0.2.0 bundle or bundle-install proof exists. This journey is verified with binaries built from its exact source commit and must not be published as release-current until platonic-v0.2.0 exists.

Platonic is the self-hosted agent server. It owns workspaces, agent profiles, threads, provider calls, tools, policy, approvals, and ledgers. Plato Agent is the client distribution built on Platonic; its commands are plato and plato-tui.

Use the site’s single working model as the map for the terms below.

Part What it means
Workspace A named, registered directory with a server-minted id and one ledger. Registration is explicit; identity is not inferred from the directory name.
Agent profile Immutable configured data bound to one workspace: model, reasoning effort, approval policy, and default toolset. It is not a process or plugin. The first-run path uses the built-in plato identity and resolved defaults, so it does not require a separate platonic agent create.
Thread A durable conversation with immutable authority: working directory, model, approval policy, and allowed effects. A thread can carry many turns and runs.
Run One submitted task driven by the server. Provider requests, tool decisions, results, and the final answer belong to that run.
Ledger The workspace-owned durable record used for status, transcript loading, and offline replay. It is the audit record, not a cache of the TUI screen.

An agent profile supplies defaults when a thread is created. The created thread records the authority it actually received; later profile or config changes do not rewrite that record. A child thread may receive a narrower toolset, path set, repository set, network grant, or approval policy than its parent, but it cannot use inheritance to widen the parent’s authority.

That split makes history interpretable: the profile answers “what defaults were offered?” and the thread answers “what was this conversation allowed to do?”

  • Closing Plato Agent leaves the host server, thread authority, completed run, and ledger in place. Reattach while that daemon is running to see the thread transcript again.
  • plato replay reads the committed ledger without a server, provider call, or tool execution.
  • Restarting the server preserves registered workspaces, agent profiles, thread authority, and completed ledger records. Live thread state is process-local, so an old thread is listed as unloaded after restart; start a new TUI thread for new work.
  • If the server stops during a run, restart recovery records the interrupted run as failed. Partial streaming text is not promoted into a completed replay answer.

Follow the first productive journey to see each boundary and its exact checkpoint.

Continue with daily operation, approvals, and providers or the Developer guide for architecture and protocol internals.