Overview

The Plext CLI installs the plext-write and plext-read skills into your AI coding agent so it can save and recall knowledge as you work.

What the CLI is for

When you code with an AI agent, valuable knowledge gets created — decisions, bug fixes, patterns, learnings. That knowledge usually lives and dies in the chat window. Plext CLI gives your agent a skill that lets it save the important bits to a Plext plex, where they become available to every future session and every other agent on the project.

The save path is explicit and agent-initiated — by design. No automatic transcript extraction. No background processes. The agent saves when it (or you) decides something is worth saving.

Supported agents

Each install is agent-native and uses the agent's own progressive-disclosure mechanism: a lean router file tells the agent when to load the full skill, which in turn references per-plex schemas and static guides on demand.

Install

# Homebrew
brew install plext-cli

# Or npm
npm install -g plext-cli

# Verify
plext --version

First run: plext setup

One command — bootstraps everything.

cd ~/code/my-project
plext setup

Performs three steps:

1. plext login — opens an OAuth flow in your browser.
2. plext link — pick a workspace, pick which plexes to capture into, pick which agents to hook up.
3. plext start — install the plext-write and plext-read skills into each chosen agent.

Once installed, ask your agent to "save this to plext" and it will. Ask it to "check our notes on X" and it will read before acting.

What gets installed where

Per-project (committed to your repo):

.plext/
├── config.json    # Workspace + plexes + agents
└── .gitignore     # Ignores local state

# Plus agent-specific skill files in .claude/, .agents/, or .cursor/

User-level (not committed):

~/.plext/
└── credentials.json  # OAuth session (0600)

A teammate cloning the repo runs plext login && plext start to pick up the same setup.

Non-interactive setup (CI / agents)

# One-command bootstrap for non-TTY environments
plext setup --api-key plxt_… --plexes decisions,incidents --agents claude-code

# Or staged via env var
export PLEXT_API_KEY=plxt_…
plext login                                    # validates key, stashes it
plext link --plexes decisions,incidents --agents ''   # '' = CLI-only mode
plext start

plext link infers the workspace from an API key's scope automatically.

Per-plex access levels

Each linked plex carries an access level in .plext/config.json. Three values:

# Bulk — every selected plex gets the same level
plext link --plexes decisions,incidents --access full

# Per-plex overrides
plext link --plexes decisions,incidents,archive \
  --access decisions=full,incidents=write,archive=read

Under an API key, access levels are capped by what the key grants on each plex.

Next steps