Skip to content

Configuration ​

Documentation

Use Settings for saved preferences and command-line flags for launch options. Do not edit Plous revision files by hand.

Command-line flags ​

FlagPurpose
--versionPrint the application version and exit
--data-dir PATHChoose a profile directory instead of ~/.plous/
--allow-execAllow agent commands in existing containers for this launch
--diagnostic-image IMAGEAllow temporary diagnostic Jobs for this launch
--helpPrint usage

For a separate profile:

sh
plous --data-dir "$HOME/plous-profiles/testing"

Profiles have separate settings, credentials, workspaces, and conversations. They can still connect to the same cluster. Only one process can open a profile at a time.

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

Settings ​

Press ss or type :s while navigating.

SettingWhere
Agent and native CLI entry commandsApplication → Agent
OpenAI and Anthropic connectionsApplication → Agent → Plous
Mouse supportApplication
Model for built-in chatType /model in the Plous composer
Workspace name and connectionsCurrent Workspace
Workspace locationLocal only. Remote is disabled.

Save applies a change. Esc cancels. Exiting does not apply an unfinished form.

Tool connections ​

A workspace can contain several cluster contexts. Kubernetes, Helm, Argo CD, and Flux share each cluster connection.

See connection fields. Saved changes appear in Activity.

Kubernetes connections ​

Saved connections keep a kubeconfig path, context, and optional default namespace. Setup starts with ~/.kube/config. After editing the path, select Refresh Contexts before adding a context.

Current Workspace uses the terminal's KUBECONFIG path list, or ~/.kube/config if unset. It follows Kubernetes merge rules. Choosing another kubeconfig in Current Workspace Settings lasts for that session.

Plous notices external context changes but does not switch an open view. Selecting a namespace in Explore does not edit kubeconfig or change another client's context.

Authentication uses the referenced kubeconfig and its plugins. Plous does not copy those credentials. See connection setup.

AI connections ​

Configure OpenAI or Anthropic under Application → Agent → Plous → Add provider connection. Saving a provider does not switch the active agent.

Type /model afresh in the Plous composer to choose a model. The selection applies across workspaces and is fixed for each response when it starts.

Claude Code and Codex use their own authentication and model settings. Edit Entry command to use a wrapper such as celai codex. Saving a different agent or command closes the native session after saving its evidence.

Provider options include a display name, API proxy, environment variable for the key, and custom instructions. A saved key takes precedence over an environment reference. Proxies require HTTPS except on loopback.

See AI setup, credentials, and active diagnostics.

Agent runtime configuration ​

Optional configuration lives at <data-dir>/agent/config.yaml. Set PLOUS_AGENT_CONFIG to use another file.

Defaults:

yaml
max_calls: 128
max_tokens: 1000000
timeout_seconds: 900

These positive limits apply to one built-in Plous response, including context compaction:

SettingLimit
max_callsModel calls
max_tokensTotal tokens used by the turn
timeout_secondsElapsed time

They do not change the model's context capacity. Plous does not automatically retry with another provider or model. Remove retired history_mode and rlm settings, which are rejected.

Native Claude Code and Codex sessions manage their own calls, context, retries, and spending. These turn limits do not apply to them.

For optional semantic history search, set both embedding.connection and embedding.model. The connection is independent of chat selection. Without embeddings, Monitor history uses lexical search. General history always uses lexical search.

Keep provider keys in Settings or an environment-variable reference, not this file.

Remembered preferences ​

Plous saves panel visibility, assistant size, tree folds, conversations, and drafts. Each connection also remembers its Explore namespace, resource type, filters, and sorting.

Each launch starts in Overview for the terminal's current context. Text searches, open resource content, unsaved edits, and chart samples are temporary.

The layout adapts to terminal size. There is no user-configurable layout or theme file. See Data and privacy for stored files.