Run your installed Claude Code or Codex CLI in the right panel. Plous connects it to cluster tools automatically.

For agents running outside Plous, use MCP setup or plous-debug.

Use the assistant panel

  • Install and sign in to   claude   or   codex  .

  • Select the cluster you want to investigate in Plous.

  • Open Settings → Application → Agent with   ss  .

  • Choose Claude Code or Codex and save.

  • Check the cluster name in the assistant header, then press   cc   to type.

The CLI starts when its panel is visible. It uses its own authentication and model settings. You do not need a Plous API key.

For a wrapper such as   celai codex  , edit Entry command in the same Settings view. See agent setup.

Use the native terminal

The CLI owns its prompt, Tab, Esc, slash commands, and conversation menus.

Key

Action

  cc   from navigation

Enter the terminal or reopen a closed session

Alt+Left / Alt+Right

Move between Plous panels

Ctrl+]

Leave the agent or return to it

Ctrl+C in the terminal

Send Ctrl+C to the CLI

Ctrl+] then Ctrl+C

Save and exit Plous

  :X   from navigation

Close the native session

  :C   from navigation

Start a new native conversation

Clicking another panel moves focus. Clicking the agent resumes the same session. Hiding or resizing the panel leaves it running.

Read earlier output

Use the wheel over the panel or Page Up / Down. Normal-screen history keeps up to 5,000 rows in memory.

Press End or Esc, scroll to the bottom, or start typing to return to the live prompt. Fullscreen CLI screens handle their own history.

Conversation storage and working directory

The CLI runs in the directory where you launched Plous. It keeps its own conversation files and credentials.

Plous associates sessions with the agent, launch directory, and cluster. Reopening can resume that conversation. Use the CLI's   /resume   for its history.

Plous saves tool results separately. It does not copy native prompts and replies into Plous chat.

Try an investigation

Select a Pod and ask:

Why is this Pod restarting? Check its status, recent events, and the failing container's logs.

Then ask:

Show the relevant logs in Plous.

You can also ask for current resource usage, a Helm revision comparison, or an Argo CD or Flux failure. Include the resource name and namespace when the current page does not identify them.

Skills, tools and MCP

MCP connects the CLI to Plous's tools. Plous supplies debugging instructions automatically, so the panel needs no separate skill installation.

Hosted sessions disable repository tools, delegation, personal skills, plugins, hooks, and unrelated MCP servers. The CLI's native sign-in and slash commands remain available. Global CLI settings stay unchanged.

Use plous-debug outside Plous when you want your agent's normal repository tools as well as cluster tools.

Available tools

Task

Examples

Read cluster data

Resources, events, logs, metrics, Helm releases, Argo CD and Flux status

Read saved evidence

Earlier tool results and workspace Activity

Open a Plous view

Resource details, lists, logs, Helm comparisons, saved evidence

The MCP reference lists tool names and limits.

Opening a view keeps your keyboard focus and draft. An unfinished form or editor can block the request.

Inspect saved evidence

  • Press Ctrl+] to leave the agent, then   :X   to close its session.

  • Select Plous in Settings.

  • Open history with   :H  .

  • Choose Claude Code tool evidence or Codex tool evidence for that connection.

  • Select a reply with   [   /   ]  , then press Enter for Sources.

Sources reads saved results without contacting the cluster again. See Sources controls.

Switch connections or close a session

A running session keeps its original cluster even when you browse another connection.

To change the agent's target:

  • Press Ctrl+], then   :X  .

  • Select the intended cluster.

  • Press   cc   and check the new session's header.

Closing the session or exiting Plous ends its MCP access. Moving panels, opening Help, or hiding the assistant does not close it.

Scope, access and limits

Cluster reads can use any namespace allowed by your Kubernetes credentials. They cannot access a different cluster. General chat and archived histories have no live tools.

Agents read by default. Start Plous with   --allow-exec   for container commands or   --diagnostic-image IMAGE   for temporary tests. These operations stay in the selected default namespace. See Active diagnostics.

Resource editing, deployment, rollback, scaling, and deletion remain user actions.

Plous limits individual tool operations. The CLI manages its model context, retries, and spending. There is no Plous-imposed session deadline.

The local connection stays on loopback, but cluster data can go to the CLI's provider. See Data and privacy.

Troubleshoot the connection

Problem

Action

Executable unavailable

Check Entry command and the PATH inherited by Plous

Sign-in required

Use the CLI's native authentication

Tools use the previous cluster

Close the session, choose the cluster, and reopen

No live tools

Select a cluster connection

Requested view blocked

Finish the form or editor and ask again

Tool access stopped

Fix the reported error, close with   :X  , and reopen with   cc

Evidence cannot be saved

Fix the storage error and retry   :X

Launch options rejected

Check the installed CLI's compatibility and the full startup error

Native launch checks used Claude Code 2.1.284 and codex-cli 0.160.0 with mock providers. Other versions must support the launch options and tool controls. There is no exact Codex version allowlist.

See MCP connection details or AI troubleshooting.

View source

On this page

Plous · Open source · Early development

Search Plous docs

Search the documentation.

Documentation versions